Research ArticleProcessing Optimization of Typed Resources with SynchronizedStorage and Computation Adaptation in Fog Computing

Zhengyang Song1 Yucong Duan 1 Shixiang Wan2 Xiaobing Sun 3 Quan Zou 2

Honghao Gao45 and Donghai Zhu1

1State Key Laboratory of Marine Resource Utilization in the South China Sea College of Information Science and TechnologyHainan University Haikou China2School of Computer Science and Engineering Tianjin University Tianjin China3School of Information Engineering Yangzhou University Yangzhou China4Computing Center Shanghai University Shanghai China5Shanghai Key Laboratory of Intelligent Manufacturing and Robotics Shanghai China

Correspondence should be addressed to Yucong Duan duanyuconghotmailcom

Received 27 January 2018 Accepted 16 April 2018 Published 30 May 2018

Academic Editor Xuyun Zhang

Copyright copy 2018 Zhengyang Song et alThis is an open access article distributed under theCreativeCommonsAttribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Wide application of the Internet of Things (IoT) system has been increasingly demanding more hardware facilities for processingvarious resources including data information and knowledge With the rapid growth of generated resource quantity it is difficultto adapt to this situation by using traditional cloud computing models Fog computing enables storage and computing services toperform at the edge of the network to extend cloud computing However there are some problems such as restricted computationlimited storage and expensive network bandwidth in Fog computing applications It is a challenge to balance the distributionof network resources We propose a processing optimization mechanism of typed resources with synchronized storage andcomputation adaptation in Fog computing In this mechanism we process typed resources in a wireless-network-based three-tier architecture consisting of Data Graph Information Graph and Knowledge GraphThe proposedmechanism aims to minimizeprocessing cost over network computation and storage while maximizing the performance of processing in a business value drivenmanner Simulation results show that the proposed approach improves the ratio of performance over user investment Meanwhileconversions between resource types deliver support for dynamically allocating network resources

1 Introduction

With the extensive development of IoT applications it is nec-essary to promote storage computing and communicationcapability of IoT devices such as efficient storage semanticintegration and parallel processing Due to powerful com-puting capability compared with edge devices cloud com-puting puts all the computing tasks on the cloud to processdata efficiently and provides highly vitalized storage andcomputing services on top of massively parallel distributedsystems [1ndash3] However with the growing quantity of datagenerated at the edge the speed is becoming the bottleneckfor the cloud-based computing paradigm [4] Also futurewireless access networks will face the challenge of transport-ing the ever-increasing volume of data centric trafficwith ever

tighter timing and quality of service (QoS) requirements [5]In order to reduce the communication bandwidth neededbetween edge devices and the central data centre nascenttechnologies and applications for mobile computing and IoTare driving computing towards dispersion [6] Conventionalmethods which are based on bandwidth allocation schemeimprove effectively bandwidth utilization efficiency [7 8]Some methods [9ndash11] achieve real-time data analysis butthey have not considered user investment [12 13] processingcost [14] and the efficient utilization of IoT resources withreduced risks [15 16] of security and robustness [17] issues

Fog computing stresses the proximity to terminal usersand client objectives dense geographical distribution andlocal resource pools latency reduction and backbone band-width savings to achieve better quality of service (QoS) [18]

HindawiWireless Communications and Mobile ComputingVolume 2018 Article ID 3794175 13 pageshttpsdoiorg10115520183794175

2 Wireless Communications and Mobile Computing

Meanwhile the computing paradigm pushes data analyticand knowledge generation power away from centralizednodes to the logical extremes of a network near the source ofthe data Therefore a large number of heterogeneous com-plexity and hierarchy resources can be processed effectivelyin fog computing systemsHowever storage and computationpower in fog computing systems are not balanced We arguethat it is a challenge to provide better load balancing ofresources in the Internet of Things From the perspective ofresource owners obtaining cost-effective resources process-ing services remains the primary concern for adoption offog computing services In [19] the authors not only studiedthe tradeoff between network performance improvement andthe communication overhead of transmitting network infor-mation but also proposed to optimize resource allocation inwireless networks

Fog devices have increasingly generated a large amountof resources including data information and knowledgeIn [20] Chaim illustrated the concepts of defining datainformation and knowledge We use Knowledge Graph tosolve the problem of extracting relationships from sourcesof knowledge [21 22] Knowledge Graph has become apowerful tool for representing knowledge in the form oflabelled digraphs and gives textual information semanticsKnowledge base contains a set of concepts examples andrelationships [23] In [24] Duan et al clarified the archi-tecture of Knowledge Graph in terms of data informationknowledge and wisdom In [25] the authors proposed todesignate the form of Knowledge Graph as four basic formsData Graph Information Graph Knowledge Graph andWisdom Graph In [26] the authors proposed to answerfive Ws questions through constructing the architecturecomposed of Data Graph Information Graph KnowledgeGraph and Wisdom Graph The development of softwareservice system can be divided into data sharing informationtransmission and knowledge creation stages in terms of datainformation and knowledge [27] In fog computing systemsthe computing resources for providing device services arehighly limited such as storage capacity computing power andnetwork bandwidth The demand for huge search space andcomputation power has become a big challenge We proposea mechanism to improve the ratio of performance overinvestment when providing resources processing servicesThe proposed approach minimizes processing cost overnetwork computation and storage while maximizing theperformance of processing in a business value drivenmannerMultitypes usersrsquo investment and corresponding benefit ratedecide final resource combination cases which makes thesystem have a relatively stable state Furthermore we designseveral simulations to verify the feasibility and evaluate theperformance of the proposed approach Currently we did notconsider negative situations of which improper arrangementof resources might lead to penalties [28] instead of rewards

The rest of this paper is organized as follows In Section 2we define typed resources and explain the proposed systemmodel Section 3 provides a running example Section 4presents practice of storage and computing collaborativeadaptation towards typed IoT resources Section 5 reports thesimulation and summarizes the simulation results Section 6

reports relatedwork Finally we draw conclusions and outlineaspects of future work in Section 7

2 Definitions and System Model

In this section we propose the definition of typed resourcesFurthermore we design the processing architecture of typedresources in the Internet of Things In order to satisfy thetransmission demands of nodes in a given wireless networkwe construct the system model through restricting networkresources

21 Definitions of Typed Resources Wide application of theInternet of Things has been acquiring a huge amount oftyped resources We classify resources into three types datainformation and knowledge [27] Data is not specified forstakeholder or machine Data is the collection of discreteelements and concepts Information is not specified forstakeholder or machine Information is conveyed throughconceptual mapping and combination of data and is usedfor interaction Knowledge is used to reason and predictunknown resources such as data information and knowl-edge Table 1 provides an explanation of typed resources ofdata information and knowledge Definitions of resourceelements and graphs are as follows

Definition 1 (resource element (REDIK)) We define resourceelement as

REDIK fl ⟨DDIK IDIKKDIK⟩ (1)

DDIK IDIK and KDIK represent data information andknowledge in DIK hierarchy respectively where DIK is anabbreviation of DDIK IDIK and KDIK

Definition 2 (graphs) In our previous work [25] we specifiedKnowledge Graph in a progressive manner as four basicforms Data Graph Information Graph Knowledge Graphand Wisdom Graph In this work we propose to specify theexisting concept of Knowledge Graph in three layers Wedefine graphs as

GraphDIK fl ⟨(DGDIK) (IGDIK) (KGDIK)⟩ (2)

DGDIK IGDIK andKGDIK representDataGraph InformationGraph and Knowledge Graph respectively DGDIK is used tomodel the temporal and spatial features of DDIK IGDIK is acombination of related DDIK IGDIK expresses the interactionand transformation of IDIK between entities in the form ofa directed graph KGDIK is a collection of statistical rulessummarized from known resources

In order to optimize resources processing it is importantto convert the types of resources in the Internet of ThingsWe give the explanations of the type conversion mechanismas follows

(a) For conversion from DDIK to IDIK in the absenceof context DDIK is meaningless We convert DDIK to IDIKthrough reorganizing DDIK The new collection of DDIKcorresponds to a different class or concept which is IDIK

Wireless Communications and Mobile Computing 3

Table 1 An explanation of three types of resources

DDIK IDIK KDIK

Semantic load Not specified for stakeholdermachine Specified for stakeholdermachine Abstracting unknown resourcesForm Discrete elements Related elements Probabilistic or categorizationUsage Identification of existence after conceptualization Communication Reasoning and predictingGraph DGDIK IGDIK KGDIK

Conversion StorageProcessing Transmission

Protection UtilizationTypedresources

Compute

The Architectureof Resources Management

Collect

Process

IoT devicesDDIK

IDIK

KDIK

DGDIK

IGDIK

KGDIK

Figure 1 The processing architecture of typed resources in IoT

(b) For conversion from DDIK to KDIK DDIK inheritssemantic relationships from standard mode and is effectivelyintegrated and reused by other applications In the conversionprocess from DDIK to KDIK we use knowledge validationtechnology to eliminate redundancy and inconsistency ofDDIK through linking DDIK sources and semantic constraintsThen we identify the most reliable DDIK to form KDIK

(c) For conversion from IDIK to KDIK IDIK is used toexpress the interaction and collaboration between entitiesThrough the classifying and abstracting interactive recordsor behaviour records related to the dynamic behaviour ofentities we obtain KDIK in the form of statistical rules Weinfer KDIK from known resources and collect necessary IDIKin the process of inference through appropriate researchtechniques such as experiments and surveys

(d) Regarding conversion from IDIK to DDIK we knowthat conversion from IDIK to DDIK is the transition fromconcept set to resource instances IDIK expresses the dynamicinteraction and collaboration between entities we obtainDDIK through observing an object at a certain time in a staticstate

(e) For conversion from KDIK to DDIK according toknowledge reasoning we establish relevant examples forextracted collection of KDIK Relationships between KDIKnodes are associated with instances in the form of attributes

(f) For conversion from KDIK to IDIK the schema-lessfeature of KGDIK makes it possible to link and utilize a richerknowledge base to help users make decisions from KDIKretrieval to IDIK creation

22 Processing Architecture of Typed Resources in the Internetof Things Generally we obtain IDIK through mining DDIKand KDIK through deducting IDIK and intelligence fromKDIK

Therefore DDIK IDIK KDIK and wisdom are the layers ofa gradual relationship The Knowledge Graph is dividedinto four levels DGDIK IGDIK KGDIK and Wisdom GraphFigure 1 shows the processing architecture of typed resourcesin IoT on the basis of DGDIK IGDIK and KGDIK In thisarchitecture graph types are allowed to convert between eachotherWeprovide the definitions ofDGDIK IGDIK andKGDIKas follows

Definition 3 (DGDIK) We define DGDIK as

DGDIK

fl collection array list stack queue tree graph (3)

DGDIK is a collection of discrete elements expressed in theform of various data structures including arrays lists stackstrees and graphs DGDIK featuring a static state is more aboutexpressing a structural relationship DGDIK is used to modelthe temporal and spatial features of DGDIK

Definition 4 (IGDIK) We define IGDIK as

IGDIK fl combination related DDIK (4)

IGDIK is a combination of related DDIK IGDIK expresses theinteraction and transformation of IDIK between entities inthe form of a directed graph IGDIK records the interactionbetween entities including both direct interaction and indi-rect interaction Also IGDIK can be expressed using multipletuples

Definition 5 (KGDIK) We define KGDIK as

KGDIK fl collection Statistic Rules (5)

4 Wireless Communications and Mobile Computing

Fog Computing

Number of

BEdge node B

Transmission of DIK packets

Transmission of DIK packetsbetween local areas

Fog node

AFog node A

A

B

DIK Packets

A

Buffer size

DIK packets Network

Bandwidth

Resources flow

Cache length

Figure 2 The transmission optimization model of typed resources in IoT (IoT TOM)

KGDIK is of free schema and expresses rich semantic rela-tionships which is conductive to have a completing map-ping towards user requirements described through naturallanguage KGDIK is a collection of statistical rules summarizedfrom known resources KGDIK further improves and perfectsthe semantic relations between the entities on the basisof DGDIK and IGDIK and then forms a semantic networkconnected by a large number of interactive relationships

23 System Model We propose a transmission optimizationmodel of typed resources in the Internet of Things whichis called IoT TOM Figure 2 shows a network composed ofmultiple wireless nodes In this given network A-B is a com-munication link between fog nodesA andBWedenote buffersize of each node asHiWe denote the number ofDIK packetsto be forwarded asNi ADIKpacket is a unit of resourcemadeinto a single package in forms of a resource combination inour proposed IoT TOM We denote the average length ofDIK packets as l We denote bandwidth between A and B asBAB We denote resource flow capacity on this link as FAB Inpractice the distribution of network resources is not balancedin fog computing applications The proposed mechanismof converting resource types achieves global optimizationto deal with this problem We use resources forwarding-waiting equilibrium and bandwidth utilization equilibrium toevaluate the performance of the proposed mechanism

231 Bandwidth Utilization Equilibrium Given a wirelessnetwork we consider utilizing bandwidth efficiently It isnecessary to dynamically allocate bandwidth Therefore weconsider transmitting DIK resource packets between IoTnodes which aims to obtain enough storage and computingresources through consuming bandwidth BEbuse denotesbandwidth utilization equilibrium BEbuse is related to the

network parameters such as bandwidth between IoT nodesand resource flow Let IRban denote the bandwidth idle ratethen BEbuse is the variance of IRban IRban and BEbuse can beillustrated as

119868119877119887119886119899 =119861119894119895 minus 119865119894119895 lowast 119897

119861119894119895 i j = 1 2 n (6)

119861119864119887119906119904119890 = 119864 (1198681198771198871198861198992) minus ([119864 (119868119877119887119886119899)]

2) (7)

In (6) Bij denotes bandwidth between IoT nodes Fij repre-sents resource flow on the link

232 Forwarding-Waiting Equilibrium Resourcesforwarding-waiting time reflects the performance ofthe proposed IoT TOM which includes two parametersforwarding-waiting rate (FRwait) and forwarding-waitingequilibrium (BEwequ) It is important to reduce the trafficload and congestion in the network in the absence of packetloss to balance resource flow LetNilowastl be the size of resourcesto be forwarded at node i then BEwequ is a variance of FRwaitBEwequ and FRwait can be illustrated as

119865119877wait =119873119894 lowast 119897119867119894

i = 1 2 n (8)

119861119864wequ = 119864 (119865119877wait2) minus ([119864 (119865119877119908119886119894119905)]

2) (9)

The cooperation between forwarding-waiting equilibriumand bandwidth utilization equilibrium enables users to utilizeIoT resources more efficiently We denote the objectivefunction for defining utilization of typed resources in theInternet of Things as F which can be illustrated as

F = 120572119861119864119887119906119904119890 + 120573119861119864wequ 120572 + 120573 = 1 (10)

Wireless Communications and Mobile Computing 5

a

b

c

D

Cloud

displaystore

processcompute

communicate

Region 1

Region 2

Region 3

a

UAV-A

Resources transmission directions in fog computing

Fog node a

A

B

C

E

Resources transmission directions using previous methods

A

Figure 3 A UAVs courier service example in fog computing

In (10) 120572 and 120573 which can be obtained through datatraining are parameters that contribute to BEbuse and BEwequrespectivelyThe smaller F is the better resource flow capacitydistribution performs

3 Running Example to Provide UAVsCourier Service

Todayrsquos unmanned aerial vehicles (UAVs) courier service isa typical IoT application in fog computing However it isdifficult to address bandwidth bottlenecks and managementcomplexity of a UAV dispatch centre For example Figure 3shows the geographical location distribution of UAVs provid-ing courier services over the air Dotted directional lines rep-resent resources transmission directions using conventionalmethods which may require expensive satellite navigationThe black directional lines display resource transmissiondirections of providing courier service using a fog computingsystem

In an ideal situation IoT resources without redundancyand inconsistency are transmitted in low user investmentsThe actual situation may not follow the ideal situation Forexample DDIK resources of (longitude and latitude) featuringredundancy and inconsistency make the scale of DDIK col-lection large Therefore if storage capacity at node UAV-A isnot enough we need to consume corresponding bandwidthto transmit these DDIK resources to other nodes whichaims to obtain storage resources We consider convertingthe resource type to IDIK or KDIK when transmitting theseDDIK resources which aims to change the scale of initialDDIKcollection Table 2 shows partial DDIK resources generated by

Table 2 Partial DDIK resources generated by UAV-A

UAV Unit DDIK (8 bit) Time range Times SpaceUAV-A (longitude latitude) 201811 900ndash1200 5000 391 Kbit

UAV-A And Table 3 shows partial IDIK resources generatedby UAV-B

For UAV-A according to DDIK resources generated fromtime t1 (201811 900) to t2 (201811 920) we sum up thattwo relationships between 119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890 and timeare longitude = k1lowasttime+b1 and latitude = k2lowasttime+b2 Infact we sum up 400 linear relationships between 119897119900119899119892119894119905119906119889119890119897119886119905119894119905119906119889119890 and time roughly Conversion from DDIK to IDIK isillustrated as

119877 (119863119863119868119870 [119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890]) 997888rarr

119868119863119868119870 [1198961 lowast 119905119894119898119890 + 1198871 1198962 lowast 119905119894119898119890 + 1198872](11)

We store these IDIK resources converted from DDIK resourcesin form of IDIK k1lowasttime+b1 k2lowasttime+b2) (k1 b1 k2 b2 time(t1 t2)) which requires storage space of 32 bits Then totalstorage space to store these IDIK resources is 32 lowast 400 = 125Kbit

Table 3 shows IDIK resources generated by UAV-BThrough counting and summarizing these rules we obtain100 KDIK resources roughly Conversion from IDIK to KDIK isillustrated as

119877 (119868119863119868119870 [119875119886119905ℎ1 119875119886119905ℎ2]) 997888rarr

119870119863119868119870 [119903119886119894119899 119889119900119908119899 119880119860119881-119861](12)

We store these KDIK resources in form of KDIK (rain downUAV-B) which requires storage space of 4 bits Then totalstorage space to store these KDIK resources is 4 lowast 250 = 097Kbit

Hence these conversions between resource types reducethe scale of collected resources In our proposed resourcetype conversion mechanism there are at most 27 lowast 27 kindsof resource combinations in each node over the networkAs shown in Figure 4 we take one resource combinationcase (DD I) (DG IG IG) as an example Conversion fromDD I into I IK aims to obtain storage resources throughconsuming computing capability In general we keep thesystem in a relatively stable state with storage and computingcollaborative adaptation

Because of complex courier tasks and geographic envi-ronment conditions of UAVs resources distribution is unbal-anced Global optimization of the UAVs wireless network cansolve this problem to a large extent We design a mechanismthat minimizes processing cost over network computationand storage while maximizing the performance of processingin a business value driven manner which is illustrated inSection 4 Figure 5 displays resources allocation results inregion 1 of the wireless network by using the proposedmechanism Initial resource combinations on UAV-A andUAV-B are as follows

Combination (UAV-A) = (119863119863119868119870 119863119863119868119870 119868119863119868119870) (119863119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 12 Kbitbandwidth (Ararr a) is 6 Kpbs and buffer size (A) is 2 Kbit

6 Wireless Communications and Mobile Computing

Table 3 Partial IDIK resources generated by UAV-B

UAV Unit IDIK (8 bit) Time range Times Space

UAV-B Path 1 (sky has rain) 201811 800 63 KbitPath 2 (rain damage UAV-B) 900ndash1200

D

D D

D

D

I

I I

I

I

K

K K

K

K

DataGraph

D

I

D

Assign valuesThe combinationof IoT resources

The combination of

D

I

I

I

I

K

Conversion

I

K

Conversion

StorageInformationGraph

KnowledgeGraph

resourcesIoT resources

stored in

I

＄＄）＋＄＄）＋

）＄）＋

）＄）＋

＋＄）＋

＋＄）＋

from ２＄）＋

GraphＭ＄）＋ resources

GraphＭ＄）＋

GraphＭ＄）＋

Figure 4 Storing resources according to the resource combination (DD I) (DG IG IG)which is one of the 27lowast27 resource combinations

Combination (UAV-B) = (119863119863119868119870 119863119863119868119870 119863119863119868119870) (119870119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 26 Kbitbandwidth (Brarr a) is 6 Kpbs and buffer size (B) is 2 Kbit

Total storage spaces at nodeUAV-B and nodeUAV-B are 5Kbit and 10Kbit respectivelyThuswe convert resource typesand transmit these resources to node a which is illustrated asfollows

Combination (UAV-A)rsquo = (119863119863119868119870 IDIK KDIK) (DGDIKIGDIK 119870119866119863119868119870) As shown in Figure 5(a) the resource scaleon UAV-A is 24 Kbit and bandwidth (A rarr a) = 4 Kpbs andbuffer size (A) = 3 Kbit

Combination (UAV-B)rsquo = (119868119863119868119870 IDIK KDIK) (DGDIKKGDIK 119868119866119863119868119870) As shown in Figure 5(b) the resource scaleon UAV-B is 23 Kbit and bandwidth (B rarr a) = 3 Kpbs andbuffer size (B) = 2 Kbit

However there are two problems in implementing thisidea First storage and computation capabilities of IoTdevices in fog computing systems are different Thus it isnecessary to obtain storage capacity by transferring typedresources Second both the performance of processingresources and user investments should be considered toimprove user investment benefit

4 Storage and Computing CollaborativeAdaptation towards IoT Resources

In fog computing systems there may be insufficient storageand computation capability at some nodes We considertransmitting IoT resources to other nodes aiming to obtain

storage and computation resources through consumingbandwidth Given a wireless network some nodes consumestorage capacity to obtain enough computing resourcesSome nodes consume computing capability to obtain enoughstorage resources In general conversion between resourcetypes provides a support for balancing the distribution ofnetwork resources in the Internet of Things

In this section we describe optimizing processingmecha-nism of typed resources with collaborative storage and com-putation adaptation We define IoT resources and resourceson GrpahDIK as follows

Definition 6 (IoT resources) We define IoT resources as atuple IR = lt IRT IRS gt IRT is the type set of IoT resourcesrepresented by a triad lt irtD irtI irtK gt IRS is the scale ofdifferent kinds of IoT resources represented by a triad lt irsDirsI irsK gt Each irs denotes the scale of resource in the formof irt

Definition 7 (resources on GrpahDIK) We define resourceson GrpahDIK as a tuple RoG = ltRGT RGSgt RGT is the typeset of resources on GrpahDIK represented by a triad lt rgtDrgtI rgtK gt RGS is the scale of different kinds of resourceson GrpahDIK represented by a triad lt rgsD rgsI rgsK gt Eachrgs denotes the scale of resources in the form of rgt

41 Computation of Resource Type Conversion Cost Consid-ering the disadvantages of conventional resources processingmethods we propose to convert both types of resources in

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 3

Table 1 An explanation of three types of resources

DDIK IDIK KDIK

Semantic load Not specified for stakeholdermachine Specified for stakeholdermachine Abstracting unknown resourcesForm Discrete elements Related elements Probabilistic or categorizationUsage Identification of existence after conceptualization Communication Reasoning and predictingGraph DGDIK IGDIK KGDIK

Conversion StorageProcessing Transmission

Protection UtilizationTypedresources

Compute

The Architectureof Resources Management

Collect

Process

IoT devicesDDIK

IDIK

KDIK

DGDIK

IGDIK

KGDIK

Figure 1 The processing architecture of typed resources in IoT

(b) For conversion from DDIK to KDIK DDIK inheritssemantic relationships from standard mode and is effectivelyintegrated and reused by other applications In the conversionprocess from DDIK to KDIK we use knowledge validationtechnology to eliminate redundancy and inconsistency ofDDIK through linking DDIK sources and semantic constraintsThen we identify the most reliable DDIK to form KDIK

(c) For conversion from IDIK to KDIK IDIK is used toexpress the interaction and collaboration between entitiesThrough the classifying and abstracting interactive recordsor behaviour records related to the dynamic behaviour ofentities we obtain KDIK in the form of statistical rules Weinfer KDIK from known resources and collect necessary IDIKin the process of inference through appropriate researchtechniques such as experiments and surveys

(d) Regarding conversion from IDIK to DDIK we knowthat conversion from IDIK to DDIK is the transition fromconcept set to resource instances IDIK expresses the dynamicinteraction and collaboration between entities we obtainDDIK through observing an object at a certain time in a staticstate

(e) For conversion from KDIK to DDIK according toknowledge reasoning we establish relevant examples forextracted collection of KDIK Relationships between KDIKnodes are associated with instances in the form of attributes

(f) For conversion from KDIK to IDIK the schema-lessfeature of KGDIK makes it possible to link and utilize a richerknowledge base to help users make decisions from KDIKretrieval to IDIK creation

22 Processing Architecture of Typed Resources in the Internetof Things Generally we obtain IDIK through mining DDIKand KDIK through deducting IDIK and intelligence fromKDIK

Therefore DDIK IDIK KDIK and wisdom are the layers ofa gradual relationship The Knowledge Graph is dividedinto four levels DGDIK IGDIK KGDIK and Wisdom GraphFigure 1 shows the processing architecture of typed resourcesin IoT on the basis of DGDIK IGDIK and KGDIK In thisarchitecture graph types are allowed to convert between eachotherWeprovide the definitions ofDGDIK IGDIK andKGDIKas follows

Definition 3 (DGDIK) We define DGDIK as

DGDIK

fl collection array list stack queue tree graph (3)

DGDIK is a collection of discrete elements expressed in theform of various data structures including arrays lists stackstrees and graphs DGDIK featuring a static state is more aboutexpressing a structural relationship DGDIK is used to modelthe temporal and spatial features of DGDIK

Definition 4 (IGDIK) We define IGDIK as

IGDIK fl combination related DDIK (4)

IGDIK is a combination of related DDIK IGDIK expresses theinteraction and transformation of IDIK between entities inthe form of a directed graph IGDIK records the interactionbetween entities including both direct interaction and indi-rect interaction Also IGDIK can be expressed using multipletuples

Definition 5 (KGDIK) We define KGDIK as

KGDIK fl collection Statistic Rules (5)

4 Wireless Communications and Mobile Computing

Fog Computing

Number of

BEdge node B

Transmission of DIK packets

Transmission of DIK packetsbetween local areas

Fog node

AFog node A

A

B

DIK Packets

A

Buffer size

DIK packets Network

Bandwidth

Resources flow

Cache length

Figure 2 The transmission optimization model of typed resources in IoT (IoT TOM)

KGDIK is of free schema and expresses rich semantic rela-tionships which is conductive to have a completing map-ping towards user requirements described through naturallanguage KGDIK is a collection of statistical rules summarizedfrom known resources KGDIK further improves and perfectsthe semantic relations between the entities on the basisof DGDIK and IGDIK and then forms a semantic networkconnected by a large number of interactive relationships

23 System Model We propose a transmission optimizationmodel of typed resources in the Internet of Things whichis called IoT TOM Figure 2 shows a network composed ofmultiple wireless nodes In this given network A-B is a com-munication link between fog nodesA andBWedenote buffersize of each node asHiWe denote the number ofDIK packetsto be forwarded asNi ADIKpacket is a unit of resourcemadeinto a single package in forms of a resource combination inour proposed IoT TOM We denote the average length ofDIK packets as l We denote bandwidth between A and B asBAB We denote resource flow capacity on this link as FAB Inpractice the distribution of network resources is not balancedin fog computing applications The proposed mechanismof converting resource types achieves global optimizationto deal with this problem We use resources forwarding-waiting equilibrium and bandwidth utilization equilibrium toevaluate the performance of the proposed mechanism

231 Bandwidth Utilization Equilibrium Given a wirelessnetwork we consider utilizing bandwidth efficiently It isnecessary to dynamically allocate bandwidth Therefore weconsider transmitting DIK resource packets between IoTnodes which aims to obtain enough storage and computingresources through consuming bandwidth BEbuse denotesbandwidth utilization equilibrium BEbuse is related to the

network parameters such as bandwidth between IoT nodesand resource flow Let IRban denote the bandwidth idle ratethen BEbuse is the variance of IRban IRban and BEbuse can beillustrated as

119868119877119887119886119899 =119861119894119895 minus 119865119894119895 lowast 119897

119861119894119895 i j = 1 2 n (6)

119861119864119887119906119904119890 = 119864 (1198681198771198871198861198992) minus ([119864 (119868119877119887119886119899)]

2) (7)

In (6) Bij denotes bandwidth between IoT nodes Fij repre-sents resource flow on the link

232 Forwarding-Waiting Equilibrium Resourcesforwarding-waiting time reflects the performance ofthe proposed IoT TOM which includes two parametersforwarding-waiting rate (FRwait) and forwarding-waitingequilibrium (BEwequ) It is important to reduce the trafficload and congestion in the network in the absence of packetloss to balance resource flow LetNilowastl be the size of resourcesto be forwarded at node i then BEwequ is a variance of FRwaitBEwequ and FRwait can be illustrated as

119865119877wait =119873119894 lowast 119897119867119894

i = 1 2 n (8)

119861119864wequ = 119864 (119865119877wait2) minus ([119864 (119865119877119908119886119894119905)]

2) (9)

The cooperation between forwarding-waiting equilibriumand bandwidth utilization equilibrium enables users to utilizeIoT resources more efficiently We denote the objectivefunction for defining utilization of typed resources in theInternet of Things as F which can be illustrated as

F = 120572119861119864119887119906119904119890 + 120573119861119864wequ 120572 + 120573 = 1 (10)

Wireless Communications and Mobile Computing 5

a

b

c

D

Cloud

displaystore

processcompute

communicate

Region 1

Region 2

Region 3

a

UAV-A

Resources transmission directions in fog computing

Fog node a

A

B

C

E

Resources transmission directions using previous methods

A

Figure 3 A UAVs courier service example in fog computing

In (10) 120572 and 120573 which can be obtained through datatraining are parameters that contribute to BEbuse and BEwequrespectivelyThe smaller F is the better resource flow capacitydistribution performs

3 Running Example to Provide UAVsCourier Service

Todayrsquos unmanned aerial vehicles (UAVs) courier service isa typical IoT application in fog computing However it isdifficult to address bandwidth bottlenecks and managementcomplexity of a UAV dispatch centre For example Figure 3shows the geographical location distribution of UAVs provid-ing courier services over the air Dotted directional lines rep-resent resources transmission directions using conventionalmethods which may require expensive satellite navigationThe black directional lines display resource transmissiondirections of providing courier service using a fog computingsystem

In an ideal situation IoT resources without redundancyand inconsistency are transmitted in low user investmentsThe actual situation may not follow the ideal situation Forexample DDIK resources of (longitude and latitude) featuringredundancy and inconsistency make the scale of DDIK col-lection large Therefore if storage capacity at node UAV-A isnot enough we need to consume corresponding bandwidthto transmit these DDIK resources to other nodes whichaims to obtain storage resources We consider convertingthe resource type to IDIK or KDIK when transmitting theseDDIK resources which aims to change the scale of initialDDIKcollection Table 2 shows partial DDIK resources generated by

Table 2 Partial DDIK resources generated by UAV-A

UAV Unit DDIK (8 bit) Time range Times SpaceUAV-A (longitude latitude) 201811 900ndash1200 5000 391 Kbit

UAV-A And Table 3 shows partial IDIK resources generatedby UAV-B

For UAV-A according to DDIK resources generated fromtime t1 (201811 900) to t2 (201811 920) we sum up thattwo relationships between 119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890 and timeare longitude = k1lowasttime+b1 and latitude = k2lowasttime+b2 Infact we sum up 400 linear relationships between 119897119900119899119892119894119905119906119889119890119897119886119905119894119905119906119889119890 and time roughly Conversion from DDIK to IDIK isillustrated as

119877 (119863119863119868119870 [119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890]) 997888rarr

119868119863119868119870 [1198961 lowast 119905119894119898119890 + 1198871 1198962 lowast 119905119894119898119890 + 1198872](11)

We store these IDIK resources converted from DDIK resourcesin form of IDIK k1lowasttime+b1 k2lowasttime+b2) (k1 b1 k2 b2 time(t1 t2)) which requires storage space of 32 bits Then totalstorage space to store these IDIK resources is 32 lowast 400 = 125Kbit

Table 3 shows IDIK resources generated by UAV-BThrough counting and summarizing these rules we obtain100 KDIK resources roughly Conversion from IDIK to KDIK isillustrated as

119877 (119868119863119868119870 [119875119886119905ℎ1 119875119886119905ℎ2]) 997888rarr

119870119863119868119870 [119903119886119894119899 119889119900119908119899 119880119860119881-119861](12)

We store these KDIK resources in form of KDIK (rain downUAV-B) which requires storage space of 4 bits Then totalstorage space to store these KDIK resources is 4 lowast 250 = 097Kbit

Hence these conversions between resource types reducethe scale of collected resources In our proposed resourcetype conversion mechanism there are at most 27 lowast 27 kindsof resource combinations in each node over the networkAs shown in Figure 4 we take one resource combinationcase (DD I) (DG IG IG) as an example Conversion fromDD I into I IK aims to obtain storage resources throughconsuming computing capability In general we keep thesystem in a relatively stable state with storage and computingcollaborative adaptation

Because of complex courier tasks and geographic envi-ronment conditions of UAVs resources distribution is unbal-anced Global optimization of the UAVs wireless network cansolve this problem to a large extent We design a mechanismthat minimizes processing cost over network computationand storage while maximizing the performance of processingin a business value driven manner which is illustrated inSection 4 Figure 5 displays resources allocation results inregion 1 of the wireless network by using the proposedmechanism Initial resource combinations on UAV-A andUAV-B are as follows

Combination (UAV-A) = (119863119863119868119870 119863119863119868119870 119868119863119868119870) (119863119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 12 Kbitbandwidth (Ararr a) is 6 Kpbs and buffer size (A) is 2 Kbit

6 Wireless Communications and Mobile Computing

Table 3 Partial IDIK resources generated by UAV-B

UAV Unit IDIK (8 bit) Time range Times Space

UAV-B Path 1 (sky has rain) 201811 800 63 KbitPath 2 (rain damage UAV-B) 900ndash1200

D

D D

D

D

I

I I

I

I

K

K K

K

K

DataGraph

D

I

D

Assign valuesThe combinationof IoT resources

The combination of

D

I

I

I

I

K

Conversion

I

K

Conversion

StorageInformationGraph

KnowledgeGraph

resourcesIoT resources

stored in

I

＄＄）＋＄＄）＋

）＄）＋

）＄）＋

＋＄）＋

＋＄）＋

from ２＄）＋

GraphＭ＄）＋ resources

GraphＭ＄）＋

GraphＭ＄）＋

Figure 4 Storing resources according to the resource combination (DD I) (DG IG IG)which is one of the 27lowast27 resource combinations

Combination (UAV-B) = (119863119863119868119870 119863119863119868119870 119863119863119868119870) (119870119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 26 Kbitbandwidth (Brarr a) is 6 Kpbs and buffer size (B) is 2 Kbit

Total storage spaces at nodeUAV-B and nodeUAV-B are 5Kbit and 10Kbit respectivelyThuswe convert resource typesand transmit these resources to node a which is illustrated asfollows

Combination (UAV-A)rsquo = (119863119863119868119870 IDIK KDIK) (DGDIKIGDIK 119870119866119863119868119870) As shown in Figure 5(a) the resource scaleon UAV-A is 24 Kbit and bandwidth (A rarr a) = 4 Kpbs andbuffer size (A) = 3 Kbit

Combination (UAV-B)rsquo = (119868119863119868119870 IDIK KDIK) (DGDIKKGDIK 119868119866119863119868119870) As shown in Figure 5(b) the resource scaleon UAV-B is 23 Kbit and bandwidth (B rarr a) = 3 Kpbs andbuffer size (B) = 2 Kbit

However there are two problems in implementing thisidea First storage and computation capabilities of IoTdevices in fog computing systems are different Thus it isnecessary to obtain storage capacity by transferring typedresources Second both the performance of processingresources and user investments should be considered toimprove user investment benefit

4 Storage and Computing CollaborativeAdaptation towards IoT Resources

In fog computing systems there may be insufficient storageand computation capability at some nodes We considertransmitting IoT resources to other nodes aiming to obtain

storage and computation resources through consumingbandwidth Given a wireless network some nodes consumestorage capacity to obtain enough computing resourcesSome nodes consume computing capability to obtain enoughstorage resources In general conversion between resourcetypes provides a support for balancing the distribution ofnetwork resources in the Internet of Things

In this section we describe optimizing processingmecha-nism of typed resources with collaborative storage and com-putation adaptation We define IoT resources and resourceson GrpahDIK as follows

Definition 6 (IoT resources) We define IoT resources as atuple IR = lt IRT IRS gt IRT is the type set of IoT resourcesrepresented by a triad lt irtD irtI irtK gt IRS is the scale ofdifferent kinds of IoT resources represented by a triad lt irsDirsI irsK gt Each irs denotes the scale of resource in the formof irt

Definition 7 (resources on GrpahDIK) We define resourceson GrpahDIK as a tuple RoG = ltRGT RGSgt RGT is the typeset of resources on GrpahDIK represented by a triad lt rgtDrgtI rgtK gt RGS is the scale of different kinds of resourceson GrpahDIK represented by a triad lt rgsD rgsI rgsK gt Eachrgs denotes the scale of resources in the form of rgt

41 Computation of Resource Type Conversion Cost Consid-ering the disadvantages of conventional resources processingmethods we propose to convert both types of resources in

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

4 Wireless Communications and Mobile Computing

Fog Computing

Number of

BEdge node B

Transmission of DIK packets

Transmission of DIK packetsbetween local areas

Fog node

AFog node A

A

B

DIK Packets

A

Buffer size

DIK packets Network

Bandwidth

Resources flow

Cache length

Figure 2 The transmission optimization model of typed resources in IoT (IoT TOM)

KGDIK is of free schema and expresses rich semantic rela-tionships which is conductive to have a completing map-ping towards user requirements described through naturallanguage KGDIK is a collection of statistical rules summarizedfrom known resources KGDIK further improves and perfectsthe semantic relations between the entities on the basisof DGDIK and IGDIK and then forms a semantic networkconnected by a large number of interactive relationships

23 System Model We propose a transmission optimizationmodel of typed resources in the Internet of Things whichis called IoT TOM Figure 2 shows a network composed ofmultiple wireless nodes In this given network A-B is a com-munication link between fog nodesA andBWedenote buffersize of each node asHiWe denote the number ofDIK packetsto be forwarded asNi ADIKpacket is a unit of resourcemadeinto a single package in forms of a resource combination inour proposed IoT TOM We denote the average length ofDIK packets as l We denote bandwidth between A and B asBAB We denote resource flow capacity on this link as FAB Inpractice the distribution of network resources is not balancedin fog computing applications The proposed mechanismof converting resource types achieves global optimizationto deal with this problem We use resources forwarding-waiting equilibrium and bandwidth utilization equilibrium toevaluate the performance of the proposed mechanism

231 Bandwidth Utilization Equilibrium Given a wirelessnetwork we consider utilizing bandwidth efficiently It isnecessary to dynamically allocate bandwidth Therefore weconsider transmitting DIK resource packets between IoTnodes which aims to obtain enough storage and computingresources through consuming bandwidth BEbuse denotesbandwidth utilization equilibrium BEbuse is related to the

network parameters such as bandwidth between IoT nodesand resource flow Let IRban denote the bandwidth idle ratethen BEbuse is the variance of IRban IRban and BEbuse can beillustrated as

119868119877119887119886119899 =119861119894119895 minus 119865119894119895 lowast 119897

119861119894119895 i j = 1 2 n (6)

119861119864119887119906119904119890 = 119864 (1198681198771198871198861198992) minus ([119864 (119868119877119887119886119899)]

2) (7)

In (6) Bij denotes bandwidth between IoT nodes Fij repre-sents resource flow on the link

232 Forwarding-Waiting Equilibrium Resourcesforwarding-waiting time reflects the performance ofthe proposed IoT TOM which includes two parametersforwarding-waiting rate (FRwait) and forwarding-waitingequilibrium (BEwequ) It is important to reduce the trafficload and congestion in the network in the absence of packetloss to balance resource flow LetNilowastl be the size of resourcesto be forwarded at node i then BEwequ is a variance of FRwaitBEwequ and FRwait can be illustrated as

119865119877wait =119873119894 lowast 119897119867119894

i = 1 2 n (8)

119861119864wequ = 119864 (119865119877wait2) minus ([119864 (119865119877119908119886119894119905)]

2) (9)

The cooperation between forwarding-waiting equilibriumand bandwidth utilization equilibrium enables users to utilizeIoT resources more efficiently We denote the objectivefunction for defining utilization of typed resources in theInternet of Things as F which can be illustrated as

F = 120572119861119864119887119906119904119890 + 120573119861119864wequ 120572 + 120573 = 1 (10)

Wireless Communications and Mobile Computing 5

a

b

c

D

Cloud

displaystore

processcompute

communicate

Region 1

Region 2

Region 3

a

UAV-A

Resources transmission directions in fog computing

Fog node a

A

B

C

E

Resources transmission directions using previous methods

A

Figure 3 A UAVs courier service example in fog computing

In (10) 120572 and 120573 which can be obtained through datatraining are parameters that contribute to BEbuse and BEwequrespectivelyThe smaller F is the better resource flow capacitydistribution performs

3 Running Example to Provide UAVsCourier Service

Todayrsquos unmanned aerial vehicles (UAVs) courier service isa typical IoT application in fog computing However it isdifficult to address bandwidth bottlenecks and managementcomplexity of a UAV dispatch centre For example Figure 3shows the geographical location distribution of UAVs provid-ing courier services over the air Dotted directional lines rep-resent resources transmission directions using conventionalmethods which may require expensive satellite navigationThe black directional lines display resource transmissiondirections of providing courier service using a fog computingsystem

In an ideal situation IoT resources without redundancyand inconsistency are transmitted in low user investmentsThe actual situation may not follow the ideal situation Forexample DDIK resources of (longitude and latitude) featuringredundancy and inconsistency make the scale of DDIK col-lection large Therefore if storage capacity at node UAV-A isnot enough we need to consume corresponding bandwidthto transmit these DDIK resources to other nodes whichaims to obtain storage resources We consider convertingthe resource type to IDIK or KDIK when transmitting theseDDIK resources which aims to change the scale of initialDDIKcollection Table 2 shows partial DDIK resources generated by

Table 2 Partial DDIK resources generated by UAV-A

UAV Unit DDIK (8 bit) Time range Times SpaceUAV-A (longitude latitude) 201811 900ndash1200 5000 391 Kbit

UAV-A And Table 3 shows partial IDIK resources generatedby UAV-B

For UAV-A according to DDIK resources generated fromtime t1 (201811 900) to t2 (201811 920) we sum up thattwo relationships between 119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890 and timeare longitude = k1lowasttime+b1 and latitude = k2lowasttime+b2 Infact we sum up 400 linear relationships between 119897119900119899119892119894119905119906119889119890119897119886119905119894119905119906119889119890 and time roughly Conversion from DDIK to IDIK isillustrated as

119877 (119863119863119868119870 [119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890]) 997888rarr

119868119863119868119870 [1198961 lowast 119905119894119898119890 + 1198871 1198962 lowast 119905119894119898119890 + 1198872](11)

We store these IDIK resources converted from DDIK resourcesin form of IDIK k1lowasttime+b1 k2lowasttime+b2) (k1 b1 k2 b2 time(t1 t2)) which requires storage space of 32 bits Then totalstorage space to store these IDIK resources is 32 lowast 400 = 125Kbit

Table 3 shows IDIK resources generated by UAV-BThrough counting and summarizing these rules we obtain100 KDIK resources roughly Conversion from IDIK to KDIK isillustrated as

119877 (119868119863119868119870 [119875119886119905ℎ1 119875119886119905ℎ2]) 997888rarr

119870119863119868119870 [119903119886119894119899 119889119900119908119899 119880119860119881-119861](12)

We store these KDIK resources in form of KDIK (rain downUAV-B) which requires storage space of 4 bits Then totalstorage space to store these KDIK resources is 4 lowast 250 = 097Kbit

Hence these conversions between resource types reducethe scale of collected resources In our proposed resourcetype conversion mechanism there are at most 27 lowast 27 kindsof resource combinations in each node over the networkAs shown in Figure 4 we take one resource combinationcase (DD I) (DG IG IG) as an example Conversion fromDD I into I IK aims to obtain storage resources throughconsuming computing capability In general we keep thesystem in a relatively stable state with storage and computingcollaborative adaptation

Because of complex courier tasks and geographic envi-ronment conditions of UAVs resources distribution is unbal-anced Global optimization of the UAVs wireless network cansolve this problem to a large extent We design a mechanismthat minimizes processing cost over network computationand storage while maximizing the performance of processingin a business value driven manner which is illustrated inSection 4 Figure 5 displays resources allocation results inregion 1 of the wireless network by using the proposedmechanism Initial resource combinations on UAV-A andUAV-B are as follows

Combination (UAV-A) = (119863119863119868119870 119863119863119868119870 119868119863119868119870) (119863119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 12 Kbitbandwidth (Ararr a) is 6 Kpbs and buffer size (A) is 2 Kbit

6 Wireless Communications and Mobile Computing

Table 3 Partial IDIK resources generated by UAV-B

UAV Unit IDIK (8 bit) Time range Times Space

UAV-B Path 1 (sky has rain) 201811 800 63 KbitPath 2 (rain damage UAV-B) 900ndash1200

D

D D

D

D

I

I I

I

I

K

K K

K

K

DataGraph

D

I

D

Assign valuesThe combinationof IoT resources

The combination of

D

I

I

I

I

K

Conversion

I

K

Conversion

StorageInformationGraph

KnowledgeGraph

resourcesIoT resources

stored in

I

＄＄）＋＄＄）＋

）＄）＋

）＄）＋

＋＄）＋

＋＄）＋

from ２＄）＋

GraphＭ＄）＋ resources

GraphＭ＄）＋

GraphＭ＄）＋

Figure 4 Storing resources according to the resource combination (DD I) (DG IG IG)which is one of the 27lowast27 resource combinations

Combination (UAV-B) = (119863119863119868119870 119863119863119868119870 119863119863119868119870) (119870119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 26 Kbitbandwidth (Brarr a) is 6 Kpbs and buffer size (B) is 2 Kbit

Total storage spaces at nodeUAV-B and nodeUAV-B are 5Kbit and 10Kbit respectivelyThuswe convert resource typesand transmit these resources to node a which is illustrated asfollows

Combination (UAV-A)rsquo = (119863119863119868119870 IDIK KDIK) (DGDIKIGDIK 119870119866119863119868119870) As shown in Figure 5(a) the resource scaleon UAV-A is 24 Kbit and bandwidth (A rarr a) = 4 Kpbs andbuffer size (A) = 3 Kbit

Combination (UAV-B)rsquo = (119868119863119868119870 IDIK KDIK) (DGDIKKGDIK 119868119866119863119868119870) As shown in Figure 5(b) the resource scaleon UAV-B is 23 Kbit and bandwidth (B rarr a) = 3 Kpbs andbuffer size (B) = 2 Kbit

However there are two problems in implementing thisidea First storage and computation capabilities of IoTdevices in fog computing systems are different Thus it isnecessary to obtain storage capacity by transferring typedresources Second both the performance of processingresources and user investments should be considered toimprove user investment benefit

4 Storage and Computing CollaborativeAdaptation towards IoT Resources

In fog computing systems there may be insufficient storageand computation capability at some nodes We considertransmitting IoT resources to other nodes aiming to obtain

storage and computation resources through consumingbandwidth Given a wireless network some nodes consumestorage capacity to obtain enough computing resourcesSome nodes consume computing capability to obtain enoughstorage resources In general conversion between resourcetypes provides a support for balancing the distribution ofnetwork resources in the Internet of Things

In this section we describe optimizing processingmecha-nism of typed resources with collaborative storage and com-putation adaptation We define IoT resources and resourceson GrpahDIK as follows

Definition 6 (IoT resources) We define IoT resources as atuple IR = lt IRT IRS gt IRT is the type set of IoT resourcesrepresented by a triad lt irtD irtI irtK gt IRS is the scale ofdifferent kinds of IoT resources represented by a triad lt irsDirsI irsK gt Each irs denotes the scale of resource in the formof irt

Definition 7 (resources on GrpahDIK) We define resourceson GrpahDIK as a tuple RoG = ltRGT RGSgt RGT is the typeset of resources on GrpahDIK represented by a triad lt rgtDrgtI rgtK gt RGS is the scale of different kinds of resourceson GrpahDIK represented by a triad lt rgsD rgsI rgsK gt Eachrgs denotes the scale of resources in the form of rgt

41 Computation of Resource Type Conversion Cost Consid-ering the disadvantages of conventional resources processingmethods we propose to convert both types of resources in

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 5

a

b

c

D

Cloud

displaystore

processcompute

communicate

Region 1

Region 2

Region 3

a

UAV-A

Resources transmission directions in fog computing

Fog node a

A

B

C

E

Resources transmission directions using previous methods

A

Figure 3 A UAVs courier service example in fog computing

In (10) 120572 and 120573 which can be obtained through datatraining are parameters that contribute to BEbuse and BEwequrespectivelyThe smaller F is the better resource flow capacitydistribution performs

3 Running Example to Provide UAVsCourier Service

Todayrsquos unmanned aerial vehicles (UAVs) courier service isa typical IoT application in fog computing However it isdifficult to address bandwidth bottlenecks and managementcomplexity of a UAV dispatch centre For example Figure 3shows the geographical location distribution of UAVs provid-ing courier services over the air Dotted directional lines rep-resent resources transmission directions using conventionalmethods which may require expensive satellite navigationThe black directional lines display resource transmissiondirections of providing courier service using a fog computingsystem

In an ideal situation IoT resources without redundancyand inconsistency are transmitted in low user investmentsThe actual situation may not follow the ideal situation Forexample DDIK resources of (longitude and latitude) featuringredundancy and inconsistency make the scale of DDIK col-lection large Therefore if storage capacity at node UAV-A isnot enough we need to consume corresponding bandwidthto transmit these DDIK resources to other nodes whichaims to obtain storage resources We consider convertingthe resource type to IDIK or KDIK when transmitting theseDDIK resources which aims to change the scale of initialDDIKcollection Table 2 shows partial DDIK resources generated by

Table 2 Partial DDIK resources generated by UAV-A

UAV Unit DDIK (8 bit) Time range Times SpaceUAV-A (longitude latitude) 201811 900ndash1200 5000 391 Kbit

UAV-A And Table 3 shows partial IDIK resources generatedby UAV-B

For UAV-A according to DDIK resources generated fromtime t1 (201811 900) to t2 (201811 920) we sum up thattwo relationships between 119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890 and timeare longitude = k1lowasttime+b1 and latitude = k2lowasttime+b2 Infact we sum up 400 linear relationships between 119897119900119899119892119894119905119906119889119890119897119886119905119894119905119906119889119890 and time roughly Conversion from DDIK to IDIK isillustrated as

119877 (119863119863119868119870 [119897119900119899119892119894119905119906119889119890 119897119886119905119894119905119906119889119890]) 997888rarr

119868119863119868119870 [1198961 lowast 119905119894119898119890 + 1198871 1198962 lowast 119905119894119898119890 + 1198872](11)

We store these IDIK resources converted from DDIK resourcesin form of IDIK k1lowasttime+b1 k2lowasttime+b2) (k1 b1 k2 b2 time(t1 t2)) which requires storage space of 32 bits Then totalstorage space to store these IDIK resources is 32 lowast 400 = 125Kbit

Table 3 shows IDIK resources generated by UAV-BThrough counting and summarizing these rules we obtain100 KDIK resources roughly Conversion from IDIK to KDIK isillustrated as

119877 (119868119863119868119870 [119875119886119905ℎ1 119875119886119905ℎ2]) 997888rarr

119870119863119868119870 [119903119886119894119899 119889119900119908119899 119880119860119881-119861](12)

We store these KDIK resources in form of KDIK (rain downUAV-B) which requires storage space of 4 bits Then totalstorage space to store these KDIK resources is 4 lowast 250 = 097Kbit

Hence these conversions between resource types reducethe scale of collected resources In our proposed resourcetype conversion mechanism there are at most 27 lowast 27 kindsof resource combinations in each node over the networkAs shown in Figure 4 we take one resource combinationcase (DD I) (DG IG IG) as an example Conversion fromDD I into I IK aims to obtain storage resources throughconsuming computing capability In general we keep thesystem in a relatively stable state with storage and computingcollaborative adaptation

Because of complex courier tasks and geographic envi-ronment conditions of UAVs resources distribution is unbal-anced Global optimization of the UAVs wireless network cansolve this problem to a large extent We design a mechanismthat minimizes processing cost over network computationand storage while maximizing the performance of processingin a business value driven manner which is illustrated inSection 4 Figure 5 displays resources allocation results inregion 1 of the wireless network by using the proposedmechanism Initial resource combinations on UAV-A andUAV-B are as follows

Combination (UAV-A) = (119863119863119868119870 119863119863119868119870 119868119863119868119870) (119863119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 12 Kbitbandwidth (Ararr a) is 6 Kpbs and buffer size (A) is 2 Kbit

6 Wireless Communications and Mobile Computing

Table 3 Partial IDIK resources generated by UAV-B

UAV Unit IDIK (8 bit) Time range Times Space

UAV-B Path 1 (sky has rain) 201811 800 63 KbitPath 2 (rain damage UAV-B) 900ndash1200

D

D D

D

D

I

I I

I

I

K

K K

K

K

DataGraph

D

I

D

Assign valuesThe combinationof IoT resources

The combination of

D

I

I

I

I

K

Conversion

I

K

Conversion

StorageInformationGraph

KnowledgeGraph

resourcesIoT resources

stored in

I

＄＄）＋＄＄）＋

）＄）＋

）＄）＋

＋＄）＋

＋＄）＋

from ２＄）＋

GraphＭ＄）＋ resources

GraphＭ＄）＋

GraphＭ＄）＋

Figure 4 Storing resources according to the resource combination (DD I) (DG IG IG)which is one of the 27lowast27 resource combinations

Combination (UAV-B) = (119863119863119868119870 119863119863119868119870 119863119863119868119870) (119870119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 26 Kbitbandwidth (Brarr a) is 6 Kpbs and buffer size (B) is 2 Kbit

Total storage spaces at nodeUAV-B and nodeUAV-B are 5Kbit and 10Kbit respectivelyThuswe convert resource typesand transmit these resources to node a which is illustrated asfollows

Combination (UAV-A)rsquo = (119863119863119868119870 IDIK KDIK) (DGDIKIGDIK 119870119866119863119868119870) As shown in Figure 5(a) the resource scaleon UAV-A is 24 Kbit and bandwidth (A rarr a) = 4 Kpbs andbuffer size (A) = 3 Kbit

Combination (UAV-B)rsquo = (119868119863119868119870 IDIK KDIK) (DGDIKKGDIK 119868119866119863119868119870) As shown in Figure 5(b) the resource scaleon UAV-B is 23 Kbit and bandwidth (B rarr a) = 3 Kpbs andbuffer size (B) = 2 Kbit

However there are two problems in implementing thisidea First storage and computation capabilities of IoTdevices in fog computing systems are different Thus it isnecessary to obtain storage capacity by transferring typedresources Second both the performance of processingresources and user investments should be considered toimprove user investment benefit

4 Storage and Computing CollaborativeAdaptation towards IoT Resources

In fog computing systems there may be insufficient storageand computation capability at some nodes We considertransmitting IoT resources to other nodes aiming to obtain

storage and computation resources through consumingbandwidth Given a wireless network some nodes consumestorage capacity to obtain enough computing resourcesSome nodes consume computing capability to obtain enoughstorage resources In general conversion between resourcetypes provides a support for balancing the distribution ofnetwork resources in the Internet of Things

In this section we describe optimizing processingmecha-nism of typed resources with collaborative storage and com-putation adaptation We define IoT resources and resourceson GrpahDIK as follows

Definition 6 (IoT resources) We define IoT resources as atuple IR = lt IRT IRS gt IRT is the type set of IoT resourcesrepresented by a triad lt irtD irtI irtK gt IRS is the scale ofdifferent kinds of IoT resources represented by a triad lt irsDirsI irsK gt Each irs denotes the scale of resource in the formof irt

Definition 7 (resources on GrpahDIK) We define resourceson GrpahDIK as a tuple RoG = ltRGT RGSgt RGT is the typeset of resources on GrpahDIK represented by a triad lt rgtDrgtI rgtK gt RGS is the scale of different kinds of resourceson GrpahDIK represented by a triad lt rgsD rgsI rgsK gt Eachrgs denotes the scale of resources in the form of rgt

41 Computation of Resource Type Conversion Cost Consid-ering the disadvantages of conventional resources processingmethods we propose to convert both types of resources in

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

6 Wireless Communications and Mobile Computing

Table 3 Partial IDIK resources generated by UAV-B

UAV Unit IDIK (8 bit) Time range Times Space

UAV-B Path 1 (sky has rain) 201811 800 63 KbitPath 2 (rain damage UAV-B) 900ndash1200

D

D D

D

D

I

I I

I

I

K

K K

K

K

DataGraph

D

I

D

Assign valuesThe combinationof IoT resources

The combination of

D

I

I

I

I

K

Conversion

I

K

Conversion

StorageInformationGraph

KnowledgeGraph

resourcesIoT resources

stored in

I

＄＄）＋＄＄）＋

）＄）＋

）＄）＋

＋＄）＋

＋＄）＋

from ２＄）＋

GraphＭ＄）＋ resources

GraphＭ＄）＋

GraphＭ＄）＋

Figure 4 Storing resources according to the resource combination (DD I) (DG IG IG)which is one of the 27lowast27 resource combinations

Combination (UAV-B) = (119863119863119868119870 119863119863119868119870 119863119863119868119870) (119870119866119863119868119870119868119866119863119868119870 119868119866119863119868119870) And the resource scale on UAV-A is 26 Kbitbandwidth (Brarr a) is 6 Kpbs and buffer size (B) is 2 Kbit

Total storage spaces at nodeUAV-B and nodeUAV-B are 5Kbit and 10Kbit respectivelyThuswe convert resource typesand transmit these resources to node a which is illustrated asfollows

Combination (UAV-A)rsquo = (119863119863119868119870 IDIK KDIK) (DGDIKIGDIK 119870119866119863119868119870) As shown in Figure 5(a) the resource scaleon UAV-A is 24 Kbit and bandwidth (A rarr a) = 4 Kpbs andbuffer size (A) = 3 Kbit

Combination (UAV-B)rsquo = (119868119863119868119870 IDIK KDIK) (DGDIKKGDIK 119868119866119863119868119870) As shown in Figure 5(b) the resource scaleon UAV-B is 23 Kbit and bandwidth (B rarr a) = 3 Kpbs andbuffer size (B) = 2 Kbit

However there are two problems in implementing thisidea First storage and computation capabilities of IoTdevices in fog computing systems are different Thus it isnecessary to obtain storage capacity by transferring typedresources Second both the performance of processingresources and user investments should be considered toimprove user investment benefit

4 Storage and Computing CollaborativeAdaptation towards IoT Resources

In fog computing systems there may be insufficient storageand computation capability at some nodes We considertransmitting IoT resources to other nodes aiming to obtain

storage and computation resources through consumingbandwidth Given a wireless network some nodes consumestorage capacity to obtain enough computing resourcesSome nodes consume computing capability to obtain enoughstorage resources In general conversion between resourcetypes provides a support for balancing the distribution ofnetwork resources in the Internet of Things

In this section we describe optimizing processingmecha-nism of typed resources with collaborative storage and com-putation adaptation We define IoT resources and resourceson GrpahDIK as follows

Definition 6 (IoT resources) We define IoT resources as atuple IR = lt IRT IRS gt IRT is the type set of IoT resourcesrepresented by a triad lt irtD irtI irtK gt IRS is the scale ofdifferent kinds of IoT resources represented by a triad lt irsDirsI irsK gt Each irs denotes the scale of resource in the formof irt

Definition 7 (resources on GrpahDIK) We define resourceson GrpahDIK as a tuple RoG = ltRGT RGSgt RGT is the typeset of resources on GrpahDIK represented by a triad lt rgtDrgtI rgtK gt RGS is the scale of different kinds of resourceson GrpahDIK represented by a triad lt rgsD rgsI rgsK gt Eachrgs denotes the scale of resources in the form of rgt

41 Computation of Resource Type Conversion Cost Consid-ering the disadvantages of conventional resources processingmethods we propose to convert both types of resources in

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 7

( )( )

a

a

A

A

( )( )

Bandwidth ( ) = 6 Kpbsrarr

Bandwidth ( rarr ) = Kpbs

Buffer size () = 2 Kbit

Buffer size () = 3 Kbit

(a)

a

a

Buffer size () = 2 Kbit

Buffer size () = 05 Kbit

()( )

B

B

( )( )

Bandwidth ( ) = Kpbsrarr

Bandwidth ( ) = Kpbsrarr

(b)

Figure 5 Resource allocation results over region 1 of the UAVs wireless network through using the proposed mechanism

Table 4 Atomic conversion cost per unit resource in IR

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119863rarr119863 119888119900119904119905119862119880119899119894119905119868119877119863rarr119868 119888119900119904119905119862119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119868rarr119863 119888119900119904119905119862119880119899119894119905119868119877119868rarr119868 119888119900119904119905119862119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119868119877119870rarr119863 119888119900119904119905119862119880119899119894119905119868119877119870rarr119868 119888119900119904119905119862119880119899119894119905119868119877119870rarr119870

Table 5 Atomic conversion cost per unit resource in ROG

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119863rarr119863 119888119900119904119905119862119880119899119894119905119877119866119863rarr119868 119888119900119904119905119862119880119899119894119905119877119866119863rarr119870119868119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119868rarr119863 119888119900119904119905119862119880119899119894119905119877119866119868rarr119868 119888119900119904119905119862119880119899119894119905119877119866119868rarr119870119870119863119868119870 119888119900119904119905119862119880119899119894119905119877119866119870rarr119863 119888119900119904119905119862119880119899119894119905119877119866119870rarr119868 119888119900119904119905119862119880119899119894119905119877119866119870rarr119870

IR and RoG First we would like to convert the types ofresources in IR which needs to invest corresponding costSpecifically we assign values from REDIK to each elementof the type set IRT Then we use these values to form thecombination case IRTrsquo = lt irtDrsquo irtIrsquo irtKrsquo gt where irtDrsquo irtIrsquoand irtKrsquo belong to DDIK IDIK KDIK As shown in Table 4we denote 119888119900119904119905119862119880119899119894119905119868119877irarrj as the atomic conversion cost perunit resource in IR Conversion cost of resource types fromIR to IRTrsquo can be illustrated as

119862119868119877 = sum119888119900119904119905119862119880119899119894119905119868119877119894rarr119895 lowast 119894119903119904119894 119894 119895 isin 119863 119868 119870 (13)

Secondwe convert the types of resources inRoG Specificallywe assign values from REDIK to each resource of the typeset RGT of RG Then we use these values to form thecombination case RGTrsquo=lt rgtD rgtI rgtK gt where rgtD rgtIand rgtK belong to DDIK IDIK KDIK As shown in Table 5 wedenote 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 as the atomic type conversion costper unit resource in RoG Conversion cost of resource typesfrom RGTrsquo to RoG can be illustrated as

119862119877119866 = sum119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 lowast 119903119892119904119894 119894 119895 isin 119863 119868 119870 (14)

42 Computation of Cost of Processing IR in RoG The pro-posed processing architecture of typed resources is adoptedto process IoT resources Cost of processing IR in RoG is

Table 6 Atomic type conversion cost of REDIK

119863119863119868119870 119868119863119868119870 119870119863119868119870119863119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119863rarr119863 119888119900119904119905119875119880119899119894119905119868119877119863rarr119868 119888119900119904119905119875119880119899119894119905119868119877119863rarr119870119868119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119868rarr119863 119888119900119904119905119875119880119899119894119905119868119877119868rarr119868 119888119900119904119905119875119880119899119894119905119868119877119868rarr119870119870119863119868119870 119888119900119904119905119875119880119899119894119905119868119877119870rarr119863 119888119900119904119905119875119880119899119894119905119868119877119870rarr119868 119888119900119904119905119875119880119899119894119905119868119877119870rarr119870

Table 7 Atomic cost of transmitting unit DDIK IDIK or KDIKresource

119863119863119868119870 119868119863119868119870 119870119863119868119870119862119905119903119886119899119904119894 119862119905119903119886119899119904119863 119862119905119903119886119899119904119868 119862119905119903119886119899119904119870

related to the resource scale As shown in Table 6 we denote119888119900119904119905119875119880119899119894119905119868119877119894rarr119895 as the atomic cost of processing unit DDIKIDIK or KDIK resource in corresponding graph resourcesThen cost of processing IR in RoG can be illustrated as

C119901119903119900 = sum(119903119892119904119895 + 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895lowast1199031198921199041015840119894) lowast 119894119903119904

119894 119895 isin 119863 119868 119870(15)

We convert the resource types in hopes of reducing thescale of resources when transferring IR in fog computingsystems aiming to balance resources load in a given wirelessnetworkWe assume that bandwidth of some nodes is enoughin a period of time We transmit IR to obtain storage andcomputation resources of other nodes through consumingbandwidth aiming to satisfy usersrsquo general requirementsAs shown in Table 7 let CtransD CtransI and CtransK indicatethe atomic cost of transmitting unit DDIK IDIK or KDIKrespectively Cost of transmitting IR can be illustrated as

C119905119903119886119899119904 = sum119862119905119903119886119899119904119894 lowast 119894119903119904119894 119894 isin 119863 119868 119870 (16)

43 Calculation of Usersrsquo Investment Benefit We optimize theperformance of processing resources through converting thetype of resources and rational allocations of infrastructuresin Internet of Things According to previous subsections wecalculate the total cost of processing IR in fog computing

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

8 Wireless Communications and Mobile Computing

Table 8 Simulation setup

No Input parameters Setup1 Number of sensor nodes 102 Number of fog nodes 43 Area of the wireless network field 500m lowast 400m4 Frequency band 24 GHz

5 Storage capability of each node(bit)

A B C D E Frarr 128M 8M 256M 8M 16M 4M

6 Data width of CPU of each node A B C D E Frarr 8 bits 16 bits 16 bits 8 bits 32 bits 8bits

7 Energy model Battery

8 SensorsA B C D E F

rarr Temperature Humidity GPSTemperature Photosensitive GPS sensors

9 Simulation time 10 minutes

applications by using our proposed mechanism Total Costdenotes the total cost of this resources processing servicewhich can be illustrated as

Total Cost = 119862119868119877 + 119862119877119866 + C119901119903119900 + C119905119903119886119899119904 (17)

Let IVuser denote inquired user investment then IVuser cor-responding to each resource combination can be illustratedas

IV119906119904119890119903 = 120601 lowast Total Cost (18)

where 120601 represents the atomic investment that can beobtained through data training Different investment pro-grams correspond to different benefit ratios (Re) Re is usedto measure the ratio of performance over investment whichis illustrated as

119877119890 =119862119905119903119886119899119904 + 119862119901119903119900

IV119906119904119890119903(19)

Then we compare 119877119890 and IV119906119904119890119903 of each program with 1198771198900and IV1199061199041198901199030 to determine whether the condition ldquo119877119890 gt1198771198900amp IV119906119904119890119903ltIV1199061199041198901199030rdquo is satisfied Let Re0 be equal to thecurrent 119877119890 when 119877119890 is greater than 1198771198900 Algorithm 1 describesthe specific process of investment driven resource processingapproach

5 Experiment

The goal of our experiments is to evaluate the proposed pro-cessing optimization approach towards typed resources Wefirst established a small wireless network as our simulationenvironment with discrete illustration [29] We observed andanalysed user investment benefit (R119890) of six nodes in thissystem Then we measured some necessary network indi-cators to illustrate effects of using the proposed mechanismto optimize resources processing Finally we compared ourproposed IoT TOM with a conventional mechanism

51 Experiment Setup As shown in Figure 6 we first estab-lished a small real wireless network with fourteen nodes

Input IR RoG IVuser0 Re0Output Theminimum Re0

For each IRTDoAssign value from REDIKCompute CIRCompute CRG Compute CproCompute Ctrans Compute Total CostCompute IVuser Compute ReIf (Re lt Re0 amp IVuseri lt IVuser0)

Re0= ReReturn Re0

Algorithm 1 CalculatingRe of each resource combination at all fognodes

These nodes consisted of different kinds of Arduino MCUsSix nodes of them were equipped with ESP8266WI-FI mod-ules and some sensors for collecting environment resourcesAs shown in Table 8 we set up some necessary inputparameters that include senor nodes and energy model [30]We deployed these nodes in four subareas in an area of 500mlowast 400m

511 Experiment Procedure We aimed to illustrate the fea-sibility of our approach through providing users resourcesprocessing services The proposed mechanism allowed eachnode to finish storage computation and transmission tasksWe assumed that a user demanded to observe environmentalresources of this real wireless network anywhere For con-venience we assigned values to some necessary parametersin the equations such as 119888119900119904119905119862119880119899119894119905119868119877irarrj 119888119900119904119905119875119880119899119894119905119868119877119894rarr119895and 119888119900119904119905119862119880119899119894119905119877119866119894rarr119895 We evaluated the performance of theproposed mechanism in terms of investment benefit and Fvalue The running time is ten minutes In our experimentwe used some rules to process collected data into threeresource collections of DDIK IDIK and KDIK For examplephotosensitive sensors collected DDIK resources of lightintensity which were 2 4 8 16 1024 We processed theseDDIK resources into IDIK resources in forms of 2x (x= [1 10])

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 9

B

F

C

E

Forwarding direction

CEdge node C

Q

W

D

Transmitting resources between local areas

Local area 1

Local area 2

S

Arduino Uno

A

Local area 3

Local area 4

Esp8266

DHT11

Fog node S

11

2

2

3

3

4 4

5

5

6 6

Figure 6 The established simulation environment

Table 9 shows an example for classifying typed resources in apractical environment

512 Statistics Analysis For convenience we only consid-ered six resource combinations to illustrate the feasibilityof the proposed mechanism We estimated IVuser and Reafter calculating CIR CRG Ctrans and Cpro As shown inFigure 7 points surrounded by the red box on the Recurves indicate our recommended resource combinationsIn these resource combinations users are able to acquireresource services of maximum processing performance withminimum investment In our experiments the proposedmechanism dynamically recommended different resourcecombinations for different network environments to meetuser demands better Figure 7 shows that No 6 of node ANo 4 of node B No 5 of node C No 2 of node D No 4 ofnode E and No 2 of node F are final recommended resourcecombinations

Furthermore we measured some necessary parametersof the proposed IoT TOM including Fij l and Ni in orderto observe the state of this wireless network As shown inTable 10 FRwait of link A-F is greater than that of other nodesStorage capacity of node F is 4 Mbit which is insufficient tostore 7Mbit IoT resourcesTherefore node F consumedmorebandwidths to obtain enough storage resources comparedwith other nodes On the whole some nodes consumedcomputing capability to obtain enough storage resources

Some nodes consumed storage capacity to obtain comput-ing resources Some nodes consumed bandwidth to obtainstorage resources In order to evaluate the performance ofIoT TOM we estimated BEbuse and BEwait which are 036 and028 respectively F value reflects resource flow capacity dis-tribution In this experiment we calculated F value accordingto (10) The calculated F value is 034 which means that theproposed mechanism effectively allocated network resourcesthrough converting resource types Table 10 shows that IRbanand FRwait of most nodes are smaller than 05 which is notthe best result in terms of network parameters intuitively Inorder to illustrate advantages of our proposed approach innext subsection we compared our proposedmechanismwitha conventional mechanism under the same resources scale

52 Comparison with Other Methods We compared theproposed mechanism with a conventional bandwidth-allocation-based mechanism We used the performance oftraditional bandwidth-allocation-based mechanism as thebaseline for comparison We modelled this simulation onthe basis of NSGA-II which is a modified version of NSGA(Nondominated Sorting Genetic Algorithm) released in 1994used to optimize multiobjectives [31] The objective functionis illustrated as

119891 = 119865 +sum119908119863119894 + 119908119868119894 + 119908119870119894 i isin 1 2 3 4 5 6 (20)

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

10 Wireless Communications and Mobile Computing

Table 9 An example for classifying typed resources in a practical environment

DDIK IDIK KDIK

Light intensity 2 4 8 16 1024 2x(x = [110]) Dim light (lt2048)Temperature 25 25 25 25 25 (119899 = 100 s) Constant temperature (100 s)

(a) (b) (c)

(d) (e) (f)

Figure 7 Estimating IVuser and Re values of considered six nodes

In (20)119908119863119894 is a parameter that contributes to the scale of irsDAnd six restrictive functions are as follows

6

sum119894=1

119908119863119894 = 1 (21)

6

sum119894=1

119908119868119894 = 1 (22)

6

sum119894=1

119908119870119894 = 1 (23)

10 le6

sum119894=1

119908119863119894 lowast 119894119903119904119863 +6

sum119894=1

119908119870119894 lowast 119894119903119904119868

+6

sum119894=1

119908119870119894 lowast 119894119903119904119870 le 30

(24)

120572 + 120573 = 1 (25)

119894119903119904119863 + 119894119903119904119868 + 119894119903119904119870 = 30 (26)

We calculated user investment benefits of these two mech-anisms Figure 8 shows the Re values per node at corre-sponding investment to illustrate benefit difference between

Table 10 Some network parameters of six links on the wirelessnetwork

Links A-B B-C C-D D-E D-F A-FBij 25 08 32 09 07 12Hi 14 35 11 10 04 23Fij 10 50 60 20 40 70l 10 20 30 10 20 30N 20 40 10 30 20 10IRban 036 046 040 058 051 032FRwait 047 019 031 028 008 061

these two methods In the simulation we assigned values toparameters in the equations such as 120572 and 120573 But the actualvalue of each parameter should be obtained through datalearning According to previous experiment results we findthat investment benefit of our proposed method is smallercompared to conventional method under the same invest-ment The results show that the proposed method performsbetter when providing resources processing services

In order to reflect the performance of processingresources in this fog computing application compared withthe conventional method we analysed F value of a givenwireless network Figure 9 shows F value of two mechanisms

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 11

Figure 8 Re values generated by different methods correspondingto user investment

Figure 9 F value of two mechanisms under different investments

under different investments The simulation results indicatethat the bandwidth-allocation-based method demands four-teen units cost to achieve objective F value F value of theproposed method is smaller compared with the conventionalmethod

We used a social network to illustrate the impact ofnetwork size on performance of IoT TOM The networkcontains 1133 nodes and 10903 edges which is from Datatang[32] We used a sample set of 862 nodes and 1932 edgesincluding five given areas This simulation was conducted ona Lenovo ThinkPad E431 desktop with a 250GHz Intel corei5 CPU and 8GRAM running the pycharm Edit onWindows10 operating system As shown in Figure 10 (6 12) representsthat the number of nodes is 6 and the number of edges is 12The results show that the bigger the scale of a given wirelessnetwork is the better the distribution of network resourcesperforms in a certain range

6 Related Work

The dynamic reconstruction of computation and storageresources not only improves the utilization of resources but

Figure 10 F values of different network scales

also simplifies management Some of the workloads thatuse common resource computing and storage technologiescan handle the current cloud system to avoid saturatedclouds [33] However the cloud computing paradigm is notsuitable to solve the problem of unbalanced distributions ofnetwork resources in the Internet of Things We argue thatit is necessary to consume bandwidth to transmit resourcesbetween IoT nodes aiming to obtain storage and computa-tion resources of other nodes in order to satisfy user demandsIn [34] the authors proposed a distributed multilevel storage(DMLS) model to solve the problems of restricted computa-tion limited storage and unstable network A wireless-basedSoftware Defined Mobile Edge Computing (SDMEC) pro-vided support for autoscaling network storage resource basedon the network demand [35] It is important to dynamicallyallocate bandwidth in fog computing application In [36]the authors proposed a bandwidth allocation scheme basedon collectable information An integrated resource allocationmethod shared the limited bandwidth among multiple wire-less body area networks (WBANs) [37] In [38] the authorsstudied the tradeoff between the latency and reliability in taskoffloading to mobile edge computing (MEC) We argue thatit is important to achieve global optimization over a wirelessnetwork in forms of converting the resource types Thebandwidths of a given wireless network are expensive in fogcomputing which makes it a challenge to reduce latency ofaccessing resources In [39] the authors proposed an efficientsoftware defined data transmission scheme (ESD-DTS) tomaximize data transmission throughput and consistentlymaintain a flowrsquos latency T Taleb et alrsquos proposed schemeenforces an autonomic creation of MEC services to allowanywhere-anytime data access with optimum Quality ofExperience (QoE) and reduced latency [40] A hierarchicalmodel composing data information knowledge andwisdomis usually represented in the form of a pyramid [41] Wedivide resources in the Internet of Things into data infor-mation and knowledge We consider converting resourcetypes to improve user investment benefit in a business valuedrivenmannerWe proposed a three-tier resource processing

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

12 Wireless Communications and Mobile Computing

architecture based on Data Graph Information Graph andKnowledgeGraph aiming at optimizing storage transfer andprocessing typed resources

7 Conclusions and Future Work

Physical services in the Internet ofThings are heterogeneouslarge-scale and resource-constrained Fog computing makesIoT devices more intelligent and saves communication band-width But distributions of network resources in fog comput-ing applications are not balanced We proposed to convertresource types to deal with this problem Based on differentresource combinations that feature a better user investmentbenefit conversions between resource types change theresource scale In order to satisfy usersrsquo general requirementswe consider transmitting typed resources to obtain enoughstorage and computing resources of other nodes throughconsuming bandwidth On the whole optimized processingof typed resources with storage and computing collaborativeadaptation effectively improves the ratio of performanceover investment Meanwhile processing resources in ourproposed three-tier architecture of Data Graph InformationGraph and Knowledge Graph optimizes spatial and tem-poral efficiency The proposed mechanism achieves globaloptimization over a wireless network to provide cost-effectiveresource processing services According to simulation resultsusers invest minimum costs to obtain maximum perfor-mances of processing IoT resources Generally our proposedmechanism keeps most fog computing systems in a stablestate which is important for improving the user experience

In the future we will continue our study of resourcesmanagement to improve performance of processing in abusiness value driven manner and conduct further experi-ments We will also explore machine learning approach tooptimize processing of resources including data informationand knowledge

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This paper is supported by Hainan Key Development Pro-gram (no ZDYF2017128) NSFC (no 61662021 no 61363007and no 61502294) and the IIOT Innovation and Develop-ment Special Foundation of Shanghai (no 2017-GYHLW-01037)

References

[1] R Buyya C S Yeo S Venugopal J Broberg and I BrandicldquoCloud computing and emerging IT platforms vision hype

and reality for delivering computing as the 5th utilityrdquo FutureGeneration Computer Systems vol 25 no 6 pp 599ndash616 2009

[2] M H U Rehman V Chang A Batool and T Y Wah ldquoBigdata reduction framework for value creation in sustainableenterprisesrdquo International Journal of Information Managementvol 36 no 6 pp 917ndash928 2016

[3] M H Ur Rehman P P Jayaraman S Ur Rehman MalikA Ur Rehman Khan and M M Gaber ldquoRedEdge A novelarchitecture for big data processing in mobile edge computingenvironmentsrdquo Journal of Sensor and Actuator Networks vol 6no 3 article no 17 2017

[4] W Shi J Cao Q Zhang Y Li and L Xu ldquoEdge ComputingVision and Challengesrdquo IEEE Internet of Things Journal vol 3no 5 pp 637ndash646 2016

[5] I Tomkos L Kazovsky and K-I Kitayama ldquoNext-generationoptical access networks Dynamic bandwidth allocationresource use optimization and QoS improvementsrdquo IEEENetwork vol 26 no 2 pp 4ndash6 2012

[6] M Satyanarayanan ldquoThe emergence of edge computingrdquo TheComputer Journal vol 50 no 1 pp 30ndash39 2017

[7] Z Ma Q Zhao and J Huang ldquoOptimizing bandwidth alloca-tion for heterogeneous traffic in IoTrdquo Peer-to-Peer Networkingand Applications vol 10 no 3 pp 610ndash621 2017

[8] Y Ito H Koga and K Iida ldquoA bandwidth reallocation schemeto improve fairness and link utilization in data center networksrdquoin Proceedings of the 13th IEEE International Conference onPervasive Computing and Communication Workshops PerComWorkshops 2016 pp 1ndash4 March 2016

[9] S K Sharma and X Wang ldquoLive Data Analytics with Collab-orative Edge and Cloud Processing in Wireless IoT NetworksrdquoIEEE Access vol 5 pp 4621ndash4635 2017

[10] K Staniec and M Habrych ldquoTelecommunication platforms fortransmitting sensor data over communication networksmdashstateof the art and challengesrdquo Sensors vol 16 no 7 article no 11132016

[11] X Wu S Zhang and A Ozgur ldquoSTAC Simultaneous Trans-mitting andAir Computing inWireless Data Center NetworksrdquoIEEE Journal on Selected Areas in Communications vol 34 no12 pp 4024ndash4034 2016

[12] C Yin and C Wang ldquoThe perturbed compound Poisson riskprocess with investment and debit interestrdquo Methodology andComputing in Applied Probability vol 12 no 3 pp 391ndash4132010

[13] C Yin and Y Wen ldquoAn extension of Paulsen-Gjessingrsquos riskmodel with stochastic return on investmentsrdquo Insurance Math-ematics and Economics vol 52 no 3 pp 469ndash476 2013

[14] P Li M Zhou and C Yin ldquoOptimal reinsurance with bothproportional andfixed costsrdquo StatisticsampProbability Letters vol106 pp 134ndash141 2015

[15] D Hua and L Zaiming ldquoOptimal reinsurance with bothproportional and fixed costsrdquoAppliedMathematics-A Journal ofChinese Universities vol 27 no 2 pp 150ndash158 2012

[16] Y Wang and C Yin ldquoApproximation for the ruin probabilitiesin a discrete time risk model with dependent risksrdquo Statistics ampProbability Letters vol 80 no 17-18 pp 1335ndash1342 2010

[17] W Sun and L Peng ldquoObserver-based robust adaptive controlfor uncertain stochastic Hamiltonian systems with state andinput delaysrdquo Lithuanian Association of Nonlinear AnalystsNonlinear Analysis Modelling and Control vol 19 no 4 pp626ndash645 2014

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

Wireless Communications and Mobile Computing 13

[18] A Brogi and S Forti ldquoQoS-aware deployment of IoT applica-tions through the fogrdquo IEEE Internet of Things Journal vol 4no 5 pp 1185ndash1192 2017

[19] J Hong and V O Li ldquoOptimal resource allocation for trans-mitting network information and data in wireless networksrdquo inProceedings of the IEEE International Conference on Communi-cations ICC rsquo10 pp 1ndash5 2010

[20] C Zins ldquoConceptual approaches for defining data informationand knowledgerdquo Journal of the Association for InformationScience and Technology vol 58 no 4 pp 479ndash493 2007

[21] Y Lin Z Liu M Sun Y Liu and X Zhu ldquoLearning entityand relation embeddings for knowledge graph completionrdquo inProceedings of the in Proceedings of the Twenty-Ninth AAAIConference on Artificial Intelligence pp 2181ndash2187 Austin Texas USA 2015

[22] J Pujara HMiao L Getoor andWCohen ldquoKnowledgeGraphIdentificationrdquo in Proceedings of the 12th International SemanticWeb Conference The Semantic Web - ISWC rsquo13 vol 8218 pp542ndash557 2013

[23] O Deshpande D S Lamba M Tourn et al ldquoBuildingmaintaining and using knowledge bases A report from thetrenchesrdquo in Proceedings of the 2013 ACM SIGMOD Conferenceon Management of Data SIGMOD rsquo13 pp 1209ndash1220 2013

[24] Y Duan L Shao G Hu Z Zhou Q Zou and Z LinldquoSpecifying architecture of knowledge graph with data graphinformation graph knowledge graph and wisdom graphrdquo inProceedings of the 15th IEEEACIS International Conference onSoftware Engineering Research Management and ApplicationsSERA 2017 pp 327ndash332 gbr June 2017

[25] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical im-plementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the IEEEInternational Conference on Software Engineering ResearchManagement and Applications pp 339ndash344 2017

[26] L Shao YDuan X SunHGaoD Zhu andWMiao ldquoAnswer-ing WhoWhen What How Why through Constructing DataGraph Information Graph Knowledge Graph and WisdomGraphrdquo in Proceedings of the The 29th International Conferenceon Software Engineering and Knowledge Engineering pp 1ndash6

[27] L Shao Y Duan X Sun Q Zou R Jing and J Lin ldquoBidi-rectional value driven design between economical planningand technical implementation based on data graph informa-tion graph and knowledge graphrdquo in Proceedings of the 15thIEEEACIS International Conference on Software EngineeringResearch Management and Applications SERA 2017 pp 339ndash344 gbr June 2017

[28] DKong L Liu andYWu ldquoBest approximation andfixed-pointtheorems for discontinuous increasingmaps in Banach latticesrdquoFixed Point Theory and Applications vol 2014 no 1 pp 1ndash102014

[29] Y Dong L Song M Wang and Y Xu ldquoCombined-penalizedlikelihood estimations with a diverging number of parametersrdquoJournal of Applied Statistics vol 41 no 6 pp 1274ndash1285 2014

[30] W Sun K Wang C Nie and X Xie ldquoEnergy-based controllerdesign of stochastic magnetic levitation systemrdquo MathematicalProblems in Engineering vol 2017 no 3 pp 1ndash6 2017

[31] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multi-objectiveoptimizationNSGA-IIrdquo inParallel ProblemSolving fromNaturePPSN VI vol 1917 of Lecture Notes in Computer Science pp849ndash858 Springer Berlin Germany 2000

[32] Datatang httpwwwdatatangcomdata796[33] S Alonso-Monsalve F Garcia-Carballeira and A Calderon

ldquoFog computing through public-resource computing and stor-agerdquo in Proceedings of the 2nd International Conference on Fogand Mobile Edge Computing FMEC 2017 pp 81ndash87 May 2017

[34] J Xing H Dai and Z Yu ldquoA distributedmulti-level model withdynamic replacement for the storage of smart edge computingrdquoJournal of Systems Architecture vol 83 pp 1ndash11 2018

[35] J Al-Badarneh Y Jararweh M Al-Ayyoub M Al-Smadi andR Fontes ldquoSoftware Defined Storage for cooperative MobileEdge Computing systemsrdquo in Proceedings of the 4th Interna-tional Conference on Software Defined Systems SDS 2017 pp174ndash179 esp May 2017

[36] Y Ito H Koga and K Iida ldquoA bandwidth allocation schemeto meet flow requirements in mobile edge computingrdquo inProceedings of the 2017 IEEE 6th International Conference onCloud Networking (CloudNet) pp 114ndash5118 September 2017

[37] D-R Chen ldquoA QoS Bandwidth Allocation Method for Coex-istence of Wireless Body Area Networksrdquo in Proceedings of the25th Euromicro International Conference on Parallel Distributedand Network-Based Processing PDP 2017 pp 251ndash254 rusMarch 2017

[38] J Liu and Q Zhang ldquoOffloading Schemes in Mobile EdgeComputing for Ultra-Reliable Low Latency CommunicationsrdquoIEEE Access vol 6 pp 12825ndash12837 2018

[39] E Kim and S Kim ldquoAn Efficient Software Defined DataTransmission Scheme based onMobile Edge Computing for theMassive IoT Environmentrdquo KSII Transactions on Internet andInformation Systems vol 12 no 2 pp 974ndash987 2018

[40] T Taleb S Dutta A Ksentini M Iqbal and H Flinck ldquoMobileedge computing potential in making cities smarterrdquo IEEECommunications Magazine vol 55 no 3 pp 38ndash43 2017

[41] J Rowley ldquoThe wisdom hierarchy Representations of theDIKW hierarchyrdquo Journal of Information Science vol 33 no 2pp 163ndash180 2007

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom