Ant Colony Optimization Algorithm to Dynamic Energy Management in Cloud …downloads.hindawi.com/journals/mpe/2017/4810514.pdf · Optimization of Energy Consumption in Cloud Data

Research ArticleAnt Colony Optimization Algorithm to Dynamic EnergyManagement in Cloud Data Center

Shanchen Pang Weiguang Zhang TongmaoMa and Qian Gao

College of Computer amp Communication Engineering China University of Petroleum (East China) Qingdao Shandong 266580 China

Correspondence should be addressed to Shanchen Pang pangscupceducn

Received 12 July 2017 Revised 25 November 2017 Accepted 4 December 2017 Published 20 December 2017

Academic Editor Anna Vila

Copyright copy 2017 Shanchen Pang et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

With the wide deployment of cloud computing data centers the problems of power consumption have become increasinglyprominent The dynamic energy management problem in pursuit of energy-efficiency in cloud data centers is investigatedSpecifically a dynamic energy management system model for cloud data centers is built and this system is composed of DVSManagementModule LoadBalancingModule andTask SchedulingModule According toTask SchedulingModule the schedulingprocess is analyzed by Stochastic Petri Net and a task-oriented resource allocation method (LET-ACO) is proposed whichoptimizes the running time of the system and the energy consumption by scheduling tasks Simulation studies confirm theeffectiveness of the proposed system model And the simulation results also show that compared to ACO Min-Min and RRscheduling strategy the proposed LET-ACO method can save up to 28 31 and 40 energy consumption while meetingperformance constraints

1 Introduction

Cloud computing is a new computing model which isdeveloped from grid computing It includes a variety of tech-nologies virtualization distributed computing and greenenergy-saving technology [1] There are mainly three typesof forms to supply resources to customers Infrastructure asa Service Platform as a Service and Software as a Service Itimplies a service-oriented architecture reducing informationtechnology overhead for the end-user great flexibility reduc-ing total cost of ownership and on-demand services [2]

With the era of big data coming computing needshave been expanding A large number of future-generationdata centers will use virtualization technology and cloudcomputing technology With the expansion of the size ofcloud data centers high energy consumption of data centersbecomes a serious problem According to the relevant dataaround 2016 the data centers with more than 100 bladechassis accounted for 60 of whole market of data centers(httpssanwen8cnp24d2OE0html) The large data cen-ters consume vast quantities of power for example a datacenter with 3000 blade chassis consumes 9000 kw per hourand electricity charges are close to 12 million dollars every

year A lot of efforts have beenmade in the industry to reducethe power need in current server platformsTherefore how tomanage the application in a cloud data center in an energy-efficient way becomes an urgent problem

Energy consumption optimization technology for distrib-uted parallel computing system includes three types resourcehibernation dynamic voltage management technology andvirtualization Resource hibernation is mainly used to reduceenergy cost of idle machines but it takes a long time to startDynamic voltage scaling (DVS) is a power managementtechnique in computer architecture where the voltage usedin a component is increased or decreased depending uponcircumstances (httpsenwikipediaorgwikiDynamic volt-age scaling) but with the decrease of voltage processorsrsquoperformance will decline Dynamic voltage and frequencyscaling (DVFS) changes the frequency and voltage of thecores scaling performance and power simultaneously

Virtualization can improve the efficiency of computersand reduce costs and energy consumption Through virtu-alization technology multiple tasks run on different virtualmachines reducing energy consumption by increasing theutilization of computer resources and reducing the numberof computers required

HindawiMathematical Problems in EngineeringVolume 2017 Article ID 4810514 10 pageshttpsdoiorg10115520174810514

2 Mathematical Problems in Engineering

However the above methods have their limitations

(1) They do not consider scheduling problems to reduceenergy consumption

(2) There are few integrated management strategies toreduce power consumption

(3) They are applicable to implemented systems or systemprototypes only

Based on the above discussions this paper tackles theproblem of dynamic energy management in data centersThe contributions of our work to existing literature can besummarized as follows

(1) An integrated management strategy is developed tosolve the power consumption and the dynamic energymanagement system model for cloud data centers isbuilt

(2) A task-oriented resource allocation method (LET-ACO) is proposed in order to adapt to the dynamicand real-time situation and solve the issue of cloudcomputing resource scheduling and decrease thepower consumption of data center

2 Related Works

21 Optimization of Energy Consumption in Cloud DataCenter A large number of data centers use virtualizationtechnology and cloud computing technologyThe integrationof traditional power management technology and virtualiza-tion technology provides new solutions to solve the powermanagement issue of cloud data center [3ndash5]

Operation and power management in virtualization aremajor challenges in cloud platform Firstly the virtualresources and physical resources managed by the virtual-ization platform are separated from each other thereforehow to implement the energy management strategy ofapplication-level virtual machine is a challenging problemCloud data center is a constantly changing platform con-sidering the requirements of isolation and independence ofvirtual machines the energy management strategy of virtualmachine should be flexible Secondly the energy consump-tion of the virtual machine cannot be measured directlyfrom the hardware most energy consumptionmodels are notaccurate

Nathuji and Schwan proposed VirtualPower approach tointegrate power management mechanisms and policies withthe virtualization technologies [6] This approach supportsthe isolated and independent operation assumed by guestvirtual machines (VMs) running on virtualized platformsand globally coordinates the effects of the diverse powermanagement policies applied by these VMs to virtualizedresources Yang et al used the open-source codes and PHPweb programming to implement a resource managementsystem with power-saving method for virtual machines in[7] They proposed a system integrated with open-sourcesoftware such as KVM and Libvirt to construct a virtualcloud management platform

High performance can be achieved in cloud computingsystems using appropriate energy management algorithmsin virtual cloud platform Beloglazov and Buyya proposednovel adaptive heuristics for dynamic consolidation of VMsbased on an analysis of historical data from the resourceusage by VMs in [8] The proposed algorithms significantlyreduce energy consumption while ensuring a high level ofadherence to the service level agreement Escheikh et alproposed workload-aware power management (PM) per-formability analysis of server-virtualized system (SVS) in[9] This modeling approach delivers a precise descriptionof different entities and features of the SVS and provideseffective support for dynamic time-based PMpolicy enablingopportunistic selection of suitable power states of powermanageable component (PMC)

Dynamic voltage scaling is a power management tech-nique in computer architecture where the voltage used in acomponent is increased or decreased depending upon cir-cumstances [10] Rossi et al proposed a novel dynamic volt-age scaling (DVS) approach for reliable and energy efficientcache memories in [11] They also developed a design explo-ration framework allowing us to evaluate several possibletrade-offs between power consumption and reliability Chenet al presented a method for dynamic voltagefrequencyscaling of networks-on-chip and last level caches inmulticoreprocessor designs where the shared resources form a singlevoltagefrequency domain in [12] These techniques reduceenergy delay product by 56 compared to a state-of-the-artprior work Moons and Verhelst generalized the DynamicVoltage Accuracy Scaling concept to pipelined structuresand quantified its energy overhead in [13] DVAS technologyincludes three parts energy saving through voltage scalingaccuracy scaling through bit width reduction and voltagescaling through bit width reduction DVAS is finally appliedto a JPEG image processing application demonstrating largesystem level gains Arroba et al proposed a DVFS policy thatreduces power consumption while preventing performancedegradation and a DVFS-aware consolidation policy thatoptimizes consumption considering the DVFS configurationthat would be necessary when mapping virtual machines tomaintain Quality of Service in [14] Han et al proposed anoff-line dynamic voltage scaling (DVS) scheme that can beintegrated with EDF which is a global real-time schedulingalgorithm for symmetric multiprocessor systems in [15]However these techniques are unsatisfactory in minimizingboth schedule length and energy consumption

Resource hibernation is the setting or prediction of theclosingsleeping time for computer processing componentsThere are many challenges for resource hibernation technol-ogy in cloud computing systems with many cloud computingresources For example the shortcomings of the traditionalscheduling strategies can lead to computer load imbalancebesides using resource hibernation technologywill obviouslyseriously affect the performance of the entire system

However the above researches discuss energy-savingmethods only from the systemhardware management theappropriate task scheduling strategies can also achieve thegoal of saving energy in cloud data centers

Mathematical Problems in Engineering 3

22 Task Scheduling Optimization Cloud computing taskscheduling refers to the allocation and management ofresources in a specific cloud environment according tocertain resource usage rules Task scheduling problems arerelated to the efficiency of all computing facilities and are ofparamount importance [16] Cloud computing task schedul-ing is a NP-complete problem it can be solved in differentmethods traditional deterministic algorithms and heuristicintelligent algorithms [17ndash25] However those methods donnot take energy consumption into account and to overcomethis limitation researchers have proposed some approachesLi et al proposed a heuristic energy-aware stochastic taskscheduling algorithm called ESTS to solve energy-efficienttask scheduling problem in [26] Zhang et al presented newenergy-efficient task scheduling algorithms for both contin-uous and discrete processor speeds with detailed simulationand analyses in [27] Changtian and Jiong adopted the geneticalgorithm to parallel find the reasonable scheduling schemein [28]

Nevertheless all these studies explored different ways ofenergy conservation in cloud data center and did not considerthe integrated management strategy Our approach exploresa set of energy-saving schemes and the task schedulingalgorithm is innovated Since the task scheduling model is aNP-complete problem in cloud computing and consideringthe supremacy of ant colony optimization algorithm (ACO)for solving task scheduling optimization in the cloud and gridenvironment [29ndash31] we select ACO in this paper to find theschema for task scheduling and achieve energy-consumptionreduction in the cloud data centers

3 System Model

The dynamic energy management system model is shownin Figure 1 This system is composed of DVS ManagementModule Load BalancingModule and Task SchedulingMod-ule The cloud data center accepts a task arrival flow thatenters a waiting queue DVS Management Module observesthe load of each machine and gives commands to optimizepower consumption and the load state of the machines andthe queue state are transferred to Load Balancing Modulewhich gets the available virtual machine resources TaskScheduling Module accepts task information and assignstasks to available virtual machines

(1) DVS Management Module DVS technologies can managethe applications in a cloud data center in an energy-efficientwayThismodulemonitors the running state of eachmachineand gives commands to optimize power consumption andthe state of machine 119894 can be calculated as follows

State (119894) = RunVM (119894)MaxVM (119894) (1)

where RunVM(119894) represents the number of VM instancesrunning onmachine 119894 andMaxVM(119894) represents themaximalnumber of VM instances a machine can host it needs toconsider the situation of physical server virtualmachine loadand so on

A threshold value 120585 (0 lt 120585 lt 1) is set to 05 WhenState(119894) is higher than 120585 the DVSManagementModule issuescommands to improve machine speed level When it is lowerthan 120585 it issues downscaling commands

(2) Load Balancing Module We calculate the load rate ofeach machine according to the status data returned bythe DVS Management Module Considering the structuralcharacteristics of cloud computing system the load rate ofmachine 119894 is used as an index of resource utilization and itscalculation method is as follows

Load (119894) = radic 1119899 119899sum119894=1

(thisMI minus sysMI)2 (2)

where thisMI represents the utilization of machine 119894 sysMIdenotes the utilization of system and 119899 is the total number ofmachines

A threshold value 120579 (0 lt 120579 lt 1) is set to 065 whenLoad(119894) is lower than 120579 virtual machines carried on machine119894 are added to standby queue

(3) Task Scheduling Module Task Scheduling Module acceptstask information and assigns tasks to available virtualmachines The proposed LET-ACO algorithm is applied totask scheduling the goal is to reduce the power consumptionof data center effectively on the premise of performanceguarantee and specific method will be shown in Sections 4and 5

4 Problem Formulation

To further analyze the problem we set up the following cloudtask scheduling model

Definition 1 A set of tasks 119879119894 = 1198791 1198792 119879119898 it indicates119898 tasks in the current queue a set of virtual machines VM119895 =VM1VM2 VM119899 it indicates 119899 available computingresources the distribution of tasks is defined as an 119898 times 119899matrix 119877119898times119899

119877119898times119899 = (11987711 11987712 sdot sdot sdot 119877111989911987721 11987722 sdot sdot sdot 1198772119899 1198771198981 1198771198982 sdot sdot sdot 119877119898119899) (3)

where 119877119894119895 is the number of 119879119894 running on VM119895

For a complex task the task system first decomposesit into several smaller tasks [32] and then allocates tasksto appropriate computing resources finally the calculationresults are summarized Task decomposition should be fol-lowed by a number of principles

(1) Independence maintaining relative independencebetween subtasks

(2) Hierarchy the tasks are decomposed in layers accord-ing to certain order parameter test object andfunction


VM1VM2

VM1VM2

VM1VM2

VMa

VMb

VMn

Waiting queueTask arrival flow

Dvs management module

Task scheduling module

Assigning tasks

DVS control

Operational status (RunVM(i)

MaxVM(i))

Available VMs

Load balancing module

VM status(thisMI sysMI)

Queue status

M1

M2

Mm

u＞ＣＭＥj)F＝ＪＯjMj Sj u＝ＪＯj uＧＧj

VMStatus(Pj ＨＯＧj RjWj U＝ＪＯj

Figure 1 Dynamic energy management system model

(3) Uniformity the granularity of subtasks is homoge-neous

(4) Similarity the decomposed subtasks are as similar aspossible

To illustrate the task scheduling process we use StochasticPetri Net (SPN) [33] to build the model The task schedulingprocess for the cloud data center is shown in Figure 2

Definition 2 A Petri Net corresponds to a 6-tuple

SPN = (119875 119879 1198651198821198720 120582) (4)

where 119875 and 119879 are disjoint sets of places and transitionsrespectively119865 belongs to (119878times119879)cup(119879times119878)119882119865 rarr 119873+ is the setof arc functions1198720 119875 rarr 0 1 2 is the initial markingand 120582 is the average implementation rate of transitions

Tables 1 and 2 present the meaning of 119875 (place) and 119879(transition)

5 Resource Preallocation

In this section we describe our optimization model for taskscheduling using LET-ACO method

51 Prediction

511 Task Execution Time PredictionModel The resources inthe cloud data centers are full of uncertainty in one aspectthe hardwarersquos abilities of different resources are different andthe CPU load and network load are varying in every minuteeven the same task has different execution time in the sameresource if it is submitted at different time in the other aspectif two different tasks were executed in two resources withthe same station the execution times are different too Theuncertainty in the two aspects makes it complex to predict ataskrsquos execution time in cloud data centers

Table 1 The meaning of place

Place Explanation119901 Sets of tasks1199011198861 1199011198862 1199011198863 119901119886119898 Tasks of decomposition1199011198871 1199011198872 1199011198873 119901119887119898 Tasks to be assigned1199011198881 119901119888119899 Computing resources1199011198891 119901119889119899 Intermediate result sets119901119890 Output result sets

Table 2 The meaning of transition

Transition Explanation119905 Segmenting tasks1199051198861 1199051198862 1199051198863 119905119886119898 Inputting tasks11990511988711 1199051198871119899 11990511988721 119905119887211989911990511988731 1199051198873119899 1199051198871198981 119905119887119898119899 Matching resources1199051198881 119905119888119899 Outputting intermediate result sets119905119889 Summary processing

According to the features of heterogeneity and dynamicchanges in the cloud computing environment Dinda pre-sented a detailed statistical analysis of the dynamic loadinformation in [34] time series analysis of the traces showsthat load is strongly correlated over time and the relationshipbetween them is almost entirely linear Therefore we designa task execution time prediction model based on linearregression model In order to enhance the reliability of theprediction model we need to make a hypothesis about theapplication of prediction

(1) The task priority being basically the same it canensure the accuracy of prediction


p t

pa1

pa2

pa3

pam

ta1

ta2

ta3

tam

pb1

pb2

pb3

pbm

tb11

tb1n

tb2n

tb21

tb31

tb3n

tbm1

tbmn

pc1 tc1

pcn tcn

pd1

pdn

td pe

Figure 2 Stochastic Petri Nets model of task scheduling

(2) Experiment with tasks of the same type (the amountof resources used by the tasks)

Since the current state of the calculated node is knownwe use the following model to predict next taskrsquos executiontime on VM119895

Time119895 (119905) = 1205720 + 1205721119880cpu119895 + 1205722119865cpu119895 + 1205723119872119895 + 1205724119878119895+ 120576 (5)

where 119880cpu119895 represents the CPU utilization 119865cpu119895 representsthe CPU basic frequency 119872119895 denotes the memory capacityand 119878119895 denotes network bandwidth 1205720 1205721 1205722 1205723 and 1205724 arethe regression coefficient 120576 represents random error

This paper uses statistical methods so we get a greatamount of data task execution time CPU utilization CPUbasic frequency memory capacity and network bandwidthto compute regression coefficient The experimental resultsshow that the average relative error of task execution time ismaintained at less than 6

512 Energy Consumption Prediction Model In the realcloud computing system an important aspect of energymanagement technology is the visibility of energy usageHowever we cannot directly obtain the energy consumptionstate data of hardware due to the existence of virtualiza-tion layer Therefore we need to build an indirect energy-consumptionmeasurementmechanism for virtual machinesThis paper employs the energy-consumption measurementmodel of virtual machine used in [35] The systemrsquos energyconsumption is mainly composed of CPU memory disk

and system-idle energy consumption it can be expressed asfollows119864sys119895 = 119864cpu119895 + 119864Mem119895 + 119864Disk119895 + 119864static119895= 120572cpu119906cpu119895 + 120572mem119906mem119895 + 120572119894119900119906disk119895 + 120574 (6)

where 119906cpu119895 denotes processor utilization 119906mem119895 representsthe number of LLC (last level cache) misses for a VM acrossall cores used by it during time period and 119906disk119895 representsthe sum of bytes read and written 120572cpu 120572mem 120572119894119900 and 120574 aremodel-specific constants

Processor utilization can be easily obtained from theprocessor usage in the operating system most processorshave the LLC count function Intel Nehalem processor pro-vides this functionality on each core tracking IO operationin Hypervisor to get the sum of bytes read and writtenAfter obtaining a series of experimental data we use linearregression with ordinary least-squares estimation to obtainparameters to establish the model The experimental resultsshow that average relative error of the systemrsquos energyconsumption is kept within 10

52 Task Scheduling Algorithm Based on LET-ACO Theheterogeneous computing platformmeets the computationaldemands of diverse tasks One of the key challenges of suchheterogeneous processor systems is effective task scheduling[30] The task scheduling problem is generally NP-completeMany real-time applications are discrete optimization prob-lems There is no polynomial time algorithm to solve combi-natorial optimization problems that are NP-hard Heuristicalgorithm can solve this problem better In this paper theimproved ant colony algorithm is adopted to solve taskscheduling problem


Ant colony algorithm which has the advantages of pos-itive feedback distributed parallel computer more robust-ness and being easy to combine with other optimizationalgorithms is a heuristic algorithm with group intelligentbionic computing method Ant colony algorithm does wellin finding out the appropriate computing resources in theunknown network topology

521 System Initialization At the initial time of the systemants are randomly placed on VM119895 which needs to provideCPU basic frequency 119875119895 the number of CPU num119895 memorycapacity 119877119895 and network bandwidth 119882119895 System initializesthe value of the pheromone on each virtual machine usingthe following equation 119887cpu 119887mem and 119887net are model-specificconstants according to the proportion of each factor 119887cpu +119887mem + 119887net = 1120591119895 (0) = 119887cpu (num119895 times 119875119895) + 119887mem119877119895 + 119887net119882119895 (7)

522The State Transfer Probability Every ant chooses virtualmachines judging by pheromone and heuristic informa-tion During the iterative process 119875119894119895119896(1199051015840) (state transitionprobability) relies on both the amount of information (vir-tual machine processing capability and energy-consumptioninformation) and the existing heuristic information on thepath Individual ants represent decomposed tasks At time 1199051015840the 119896th ant chooses VM119895 for the next task 119894 with a probabilityas shown in the following equation

119875119895119896 (1199051015840) = [120591119895 (1199051015840)]120572 [120578119895]sum119899119897=1 [120591119897 (1199051015840)]120572 [120578119897] if 119897 119895 isin 1 1198990 otherwise

(8)

where 120572 is the information heuristic factor that indicates theimportance of the resource pheromone 120591119895(1199051015840) represents thepheromone concentration of VM119895 at time 1199051015840 120578119895 represents theexpectation that next taskwill be executed onVM119895 120578119895 = 119868119895(119905)and 119868119895(119905) is the profit function that represents VM119895 profitwhen next task is running on VM119895 its calculation methodis as follows 119868119895 (119905) = 1120582Time119895 (119905) minus 1205821015840119864119895 (119905) (9)

where 120582 and 1205821015840 are set according to the experimental datathen our algorithm is used to get the experimental results andthe parameters are adjusted 119875119895119896(1199051015840) is to be zero when thevirtual machine 119895 has been selected or the virtual machine119895 fails523 Pheromone Updating Pheromone updating is doneby all the ants that come up with feasible schedules in thefollowing manner as shown in the following equation120591119895 (1199051015840 + 1) = (1 minus 120588) 120591119895 (1199051015840) + Δ120591119895 (10)

where 120591119895(1199051015840 + 1) represents the pheromone concentration ofVM119895 at the next moment 120588 is volatile factor 0 le 120588 lt 1 Δ120591119895

represents pheromone increment and its calculationmethodis as follows Δ120591119895 = 119863 sdotmax (RES (119868119896119903)) (11)

where 119863 is the intensity of pheromone and it affects therate of convergence RES(119868119896119903) represents the task assignmentscheme in which the 119896th ant has searched in the 119903th iterationwhose value is determined by the profit function value of theallocation scheme

If the ant 119896 completes the round search (a task allocationscheme has been found) all the virtual machines on thispath will update the local pheromone If all ants complete theround search finding the optimal path in this iteration thevirtual machines on the optimal path will update the localpheromone

524 LET-ACO Algorithm Flow The proposed LET-ACOalgorithm is described according to the following steps

Step 1 DVSManagement Module obtains the running statusinformation of each physical host and virtual machineThenit uses DVS technology to control hosts according to State(119894)Step 2 The load state of machines and the queue state aretransferred to Load Balancing Module which can get theavailable virtual machine resources

Step 3 The profit matrix is obtained by calculating 119868119895(119905)Step 4 Task Scheduling Module sets initial pheromone foravailable computing nodes

Step 5 Initialize the ants and place them randomly on theavailable virtual machines

Step 6 Calculate the transition probability of ant 119896 accordingto the profit matrix and choose next node

Step 7 If the ant 119896 completes the round search localpheromone will be updated if not return to Step 6

Step 8 If all ants complete the round search globalpheromone will be updated if not return to Step 5

Step 9 If the number of iterations is required Task Schedul-ing Module outputs the optimal allocation scheme if notreturn to Step 5

Step 10 Determine whether there is any task to be allocatedin the task queue and if so return to Step 1 and if not endthis task assignment

LET-ACO runs periodically Figure 3 depicts the mainflow of the algorithm

6 Experiments and Performance Analysis

To evaluate our proposed method the simulation experi-ments are operated on CloudSim to analyse the effects We


Start

Getting the status information and controllingthe host voltage

Getting a list of virtual machines

Calculating the profit matrix

Initializing the virtual machine pheromone

Placing ants randomly on the available VMs

Selecting next hop

Whether ant k completes search

Calculating the transition probability of ant k

Local pheromone update

Whether all antscomplete search

Initialing ants

No

Global pheromone update

Yes

Whether the iterations meet requirements

Outputting the optimal allocation scheme

Yes

Whether there is anytask to be allocated

End

NoYes

yes

No

No

Figure 3 Main flow of the LET-ACO algorithm


Table 3 Parameters of CloudSim

Type Parameter Value

Data center Number of data centers 1Number of hosts per data center 6

Virtual machine (VM)

Total number of VMs 30MIPS of PE 500ndash1500 (MIPS)

Number of PEs per VM 2ndash8VMmemory 512ndash2048 (MB)Bandwidth 500ndash1000 bit

Task Total number of tasks 100ndash400Length of task 5000 (MI)

Table 4 Parameters of LET-ACO algorithm

Parameter ValueNumber of ants in colony 25Number of iterations 20120572 02120588 05

LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks

Figure 4 Execution times of tasks

design the simulation environment by assuming that thereare 6 hosts 30VMs and 100ndash400 tasks Data and informationabout hosts VMs and tasks are summarized in Table 3

In this experiment the parameters of the LET-ACOalgorithm are set in Table 4

Based on the parametersrsquo settings in Tables 3 and 4 resultsof three metrics are obtained and illustrated in Figures 4ndash7

Figure 4 shows the comparison of task total runningtime using LET-ACO ACO Min-Min and Round-Robin(RR) algorithms ACO algorithm [36] is the basic ant colonyalgorithm involving time factor Min-Min algorithm [37] andRR algorithm [38] are the classic algorithms employed byprocess and network schedulers in computing It is shownthat the task completion time gets longer with the increaseof tasks number Task execution time is the longest by usingRR algorithm ACO and LET-ACO algorithm are obviouslysuperior to the other algorithms These results suggest that

LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks

Figure 5 Energy consumption of tasks running in system

100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)

Figure 6 Tasksrsquo average waiting time

LET-ACO is very effective in finishing the tasks with smallVM time and it takes power into consideration but it doesnot bring any increase of the makespan

In Figure 5 it is shown that the energy that the datacenter consumes increases with the number of tasks addedCompared with the other three algorithms the proposed


LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load

algorithm can significantly reduce the systemrsquos energy con-sumption To analyze the reason it is just because ACOMin-Min and RR algorithms focus solely on the completion timeof the task without considering energy consumption

Figure 6 shows the task average waiting time for differentalgorithmsThe results show that the task waiting time basedon LET-ACO algorithm is shorter which can meet the needsof users better

Figure 7 shows the host load using different algorithmsThe cloud system load has been maintained in a relativelybalanced state by using LET-ACO algorithm Min-Min algo-rithm can complete the task in a short time but the loadbalancing of Min-Min algorithm is the worst The resourceswith strong processing ability are always in the workingstate while other resources are idle all the time Min-Minalgorithm cannot reflect the advantages of distributed system

Taken together LET-ACO algorithm can reduce thepower consumption of data center effectively on the premiseof performance guarantee

7 Conclusion

This work proposes an integrated management strategy tosolve the data center power consumption problem Theresource scheduling system model is built for cloud datacenter and this system is analysed by Stochastic Petri NetA task-oriented resource allocation method (LET-ACO) isproposed it can reduce the power consumption of datacenter effectively on the premise of performance guaranteeTo validate the effectiveness of our proposed method thesimulation experiments are operated on CloudSim

In the future work we will try to build more detailedsystem model to describe the data center or to build a newkind of model which can be greatly effective for analysingparticular problems in cloud data center including heteroge-neous tasks scheduling and fault diagnosing andwemay takemore factors into consideration for example not only timeand power consumption but also the state of the hosts caninfluence the energy of the data center another promising

future work direction is to try to use other biocomputingmethods to solve some problems in cloud data center

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was partially supported by the National Nat-ural Science Foundation of China (Grants nos 6157252261572523 and 61402533) and special fund project for workmethod innovation of Ministry of Science and Technology ofChina (no 2015IM010300)

References

[1] C Weinhardt W A Anandasivam B Blau et al ldquoCloudcomputingmdasha classification business models and researchdirectionsrdquo Business amp Information Systems Engineering vol 1no 5 pp 391ndash399 2009

[2] M Vouk ldquoCloud computing issues research and implementa-tionsrdquo Journal of Computing and Information Technology vol16 no 4 pp 235ndash246 2008

[3] K-J Ye Z-H Wu X-H Jiang and Q-M He ldquoPower man-agement of virtualized cloud computing platformrdquo JisuanjiXuebaoChinese Journal of Computers vol 35 no 6 pp 1262ndash1285 2012

[4] Y Zhang W Pan Q Wang K Bai and M Yu ldquoHypebiosenforcing vm isolation with minimized and decomposed cloudtcbrdquo Tech Rep Virginia Commonwealth University 2012

[5] Y ZhangM Li BWilderM Yu K Bai and P Liu ldquoNeuCloudenabling privacy-preserving monitoring in cloud computingrdquo

[6] R Nathuji and K Schwan ldquoVirtualPower coordinated powermanagement in virtualized enterprise systemsrdquo in Proceedingsof the SOSPrsquo07 21st ACM Symposium on Operating SystemsPrinciples pp 265ndash278 October 2007

[7] C-T Yang J-C Liu S-T Chen and K-L Huang ldquoVirtualmachine management system based on the power saving algo-rithm in cloudrdquo Journal of Network and Computer Applicationsvol 80 pp 165ndash180 2017

[8] A Beloglazov and R Buyya ldquoOptimal online deterministicalgorithms and adaptive heuristics for energy and performanceefficient dynamic consolidation of virtual machines in Clouddata centersrdquo Concurrency and Computation Practice andExperience vol 24 no 13 pp 1397ndash1420 2012

[9] M Escheikh K Barkaoui and H Jouini ldquoVersatile workload-aware power management performability analysis of servervirtualized systemsrdquo The Journal of Systems and Software vol125 pp 365ndash379 2017

[10] Y Xia M Zhou X Luo S Pang and Q Zhu ldquoA stochasticapproach to analysis of energy-aware DVS-enabled cloud dat-acentersrdquo IEEE Transactions on Systems Man and CyberneticsSystems vol 45 no 1 pp 73ndash83 2015

[11] D Rossi V Tenentes S Khursheed and B M Al-HashimildquoBTI and leakage aware dynamic voltage scaling for reliablelow power cache memoriesrdquo in Proceedings of the 21st IEEEInternational On-Line Testing Symposium IOLTS 2015 pp 194ndash199 July 2015


[12] X Chen Z Xu H Kim et al ldquoDynamic voltage and frequencyscaling for shared resources in multicore processor designsrdquo inProceedings of the 50th Annual Design Automation ConferenceDAC 2013 p 114 June 2013

[13] B Moons and M Verhelst ldquoDVAS dynamic voltage accuracyscaling for increased energy-efficiency in approximate com-putingrdquo in Proceedings of the 20th IEEEACM InternationalSymposium on Low Power Electronics and Design ISLPED 2015pp 237ndash242 July 2015

[14] P Arroba J M Moya J L Ayala and R Buyya ldquoDynamicVoltage and Frequency Scaling-aware dynamic consolidationof virtual machines for energy efficient cloud data centersrdquoConcurrency Computation Practice and Experience vol 29 no10 Article ID e4067 2017

[15] S Han M Park X Piao andM Park ldquoA dual speed scheme fordynamic voltage scaling on real-time multiprocessor systemsrdquoThe Journal of Supercomputing vol 71 no 2 pp 574ndash590 2015

[16] F Ramezani J Lu and F K Hussain ldquoTask-based system loadbalancing in cloud computing using particle swarm optimiza-tionrdquo International Journal of Parallel Programming vol 42 no5 pp 739ndash754 2014

[17] E Pacini C Mateos and C G Garino ldquoDistributed jobscheduling based on swarm intelligence a surveyrdquo Computersand Electrical Engineering vol 40 no 1 pp 252ndash269 2014

[18] T Ma Q Yan W Liu D Guan and S Lee ldquoGrid taskscheduling algorithm reviewrdquo IETE Technical Review vol 28no 2 pp 158ndash167 2011

[19] S Pang T Ma and T Liu ldquoAn improved ant colony opti-mization with optimal search library for solving the travelingsalesman problemrdquo Journal of Computational and TheoreticalNanoscience vol 12 no 7 pp 1440ndash1444 2015

[20] T Song and L Pan ldquoSpiking neural P systems with requestrulesrdquo Neurocomputing vol 193 pp 193ndash200 2016

[21] T Song F Gong X Liu Y Zhao and X Zhang ldquoSpikingneural P systems with white hole neuronsrdquo IEEE Transactionson NanoBioscience vol 15 no 7 pp 666ndash673 2016

[22] T Song P Zheng M L D Wong and X Wang ldquoDesign oflogic gates using spiking neural P systems with homogeneousneurons and astrocytes-like controlrdquo Information Sciences vol372 pp 380ndash391 2016

[23] X Wang T Song P Zheng S Hao and T Ma ldquoSpiking neuralP systems with anti-spikes and without annihilating priorityrdquoScience and Technology vol 20 no 1 pp 32ndash41 2017

[24] Y Wen H Xu and J Yang ldquoA heuristic-based hybrid genetic-variable neighborhood search algorithm for task schedulingin heterogeneous multiprocessor systemrdquo Information Sciencesvol 181 no 3 pp 567ndash581 2011

[25] X Zuo G Zhang and W Tan ldquoSelf-adaptive learning pso-based deadline constrained task scheduling for hybrid iaascloudrdquo IEEE Transactions on Automation Science and Engineer-ing vol 11 no 2 pp 564ndash573 2014

[26] K Li X Tang and K Li ldquoEnergy-efficient stochastic taskscheduling on heterogeneous computing systemsrdquo IEEE Trans-actions on Parallel and Distributed Systems vol 25 no 11 pp2867ndash2876 2014

[27] L M Zhang K Li D C-T Lo and Y Zhang ldquoEnergy-efficienttask scheduling algorithms on heterogeneous computers withcontinuous and discrete speedsrdquo Sustainable Computing Infor-matics and Systems vol 3 no 2 pp 109ndash118 2013

[28] Y Changtian and Y Jiong ldquoEnergy-aware genetic algorithmsfor task scheduling in cloud computingrdquo in ChinaGrid AnnualConference (ChinaGrid) pp 43ndash48 2012

[29] W-T Wen C-D Wang D-S Wu and Y-Y Xie ldquoAn ACO-based scheduling strategy on load balancing in cloud com-puting environmentrdquo in Proceedings of the 9th InternationalConference on Frontier of Computer Science and TechnologyFCST 2015 pp 364ndash369 August 2015

[30] H Sun and J Zhu ldquoDesign of task-resource allocation modelbased on Q-learning and double orientation aco algorithm forcloud computingrdquo Computer Measurement amp Control 2014

[31] G Lin Y Bie M Lei and K Zheng ldquoACO-BTM a behaviortrust model in cloud computing environmentrdquo InternationalJournal of Computational Intelligence Systems vol 7 no 4 pp785ndash795 2014

[32] C Mateos A Zunino andM Campo ldquoA survey on approachesto gridificationrdquo Software Practice and Experience vol 38 no5 pp 523ndash556 2008

[33] H He S Pang and Z Zhao ldquoDynamic scalable stochasticpetri net a novel model for designing and analysis of resourcescheduling in cloud computingrdquo Scientific Programming vol2016 Article ID 9259248 13 pages 2016

[34] P A Dinda ldquoThe statistical properties of host loadrdquo ScientificProgramming vol 7 no 3-4 pp 211ndash229 1999

[35] A Kansal F Zhao J Liu N Kothari and A A BhattacharyaldquoVirtual machine power metering and provisioningrdquo in Pro-ceedings of the 1st ACM Symposium on Cloud Computing (SoCCrsquo10) pp 39ndash50 June 2010

[36] C Mateos E Pacini and C G Garino ldquoAn ACO-inspiredalgorithm for minimizing weighted flowtime in cloud-basedparameter sweep experimentsrdquo Advances in Engineering Soft-ware vol 56 pp 38ndash50 2013

[37] T D Braunt H J Siegel N Beck et al ldquoA comparison study ofeleven static heuristics for mapping a class of independent tasksonto ileterogeneous distributed computing systemsrdquo 2000

[38] R H Arpaci-Dusseau and A C Arpaci-Dusseau OperatingSystems Three Easy Pieces Arpaci-Dusseau Books 2015

Submit your manuscripts athttpswwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Journal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


However the above methods have their limitations

(1) They do not consider scheduling problems to reduceenergy consumption

(2) There are few integrated management strategies toreduce power consumption

(3) They are applicable to implemented systems or systemprototypes only

Based on the above discussions this paper tackles theproblem of dynamic energy management in data centersThe contributions of our work to existing literature can besummarized as follows

(1) An integrated management strategy is developed tosolve the power consumption and the dynamic energymanagement system model for cloud data centers isbuilt

(2) A task-oriented resource allocation method (LET-ACO) is proposed in order to adapt to the dynamicand real-time situation and solve the issue of cloudcomputing resource scheduling and decrease thepower consumption of data center

2 Related Works

21 Optimization of Energy Consumption in Cloud DataCenter A large number of data centers use virtualizationtechnology and cloud computing technologyThe integrationof traditional power management technology and virtualiza-tion technology provides new solutions to solve the powermanagement issue of cloud data center [3ndash5]

Operation and power management in virtualization aremajor challenges in cloud platform Firstly the virtualresources and physical resources managed by the virtual-ization platform are separated from each other thereforehow to implement the energy management strategy ofapplication-level virtual machine is a challenging problemCloud data center is a constantly changing platform con-sidering the requirements of isolation and independence ofvirtual machines the energy management strategy of virtualmachine should be flexible Secondly the energy consump-tion of the virtual machine cannot be measured directlyfrom the hardware most energy consumptionmodels are notaccurate

Nathuji and Schwan proposed VirtualPower approach tointegrate power management mechanisms and policies withthe virtualization technologies [6] This approach supportsthe isolated and independent operation assumed by guestvirtual machines (VMs) running on virtualized platformsand globally coordinates the effects of the diverse powermanagement policies applied by these VMs to virtualizedresources Yang et al used the open-source codes and PHPweb programming to implement a resource managementsystem with power-saving method for virtual machines in[7] They proposed a system integrated with open-sourcesoftware such as KVM and Libvirt to construct a virtualcloud management platform

High performance can be achieved in cloud computingsystems using appropriate energy management algorithmsin virtual cloud platform Beloglazov and Buyya proposednovel adaptive heuristics for dynamic consolidation of VMsbased on an analysis of historical data from the resourceusage by VMs in [8] The proposed algorithms significantlyreduce energy consumption while ensuring a high level ofadherence to the service level agreement Escheikh et alproposed workload-aware power management (PM) per-formability analysis of server-virtualized system (SVS) in[9] This modeling approach delivers a precise descriptionof different entities and features of the SVS and provideseffective support for dynamic time-based PMpolicy enablingopportunistic selection of suitable power states of powermanageable component (PMC)

Dynamic voltage scaling is a power management tech-nique in computer architecture where the voltage used in acomponent is increased or decreased depending upon cir-cumstances [10] Rossi et al proposed a novel dynamic volt-age scaling (DVS) approach for reliable and energy efficientcache memories in [11] They also developed a design explo-ration framework allowing us to evaluate several possibletrade-offs between power consumption and reliability Chenet al presented a method for dynamic voltagefrequencyscaling of networks-on-chip and last level caches inmulticoreprocessor designs where the shared resources form a singlevoltagefrequency domain in [12] These techniques reduceenergy delay product by 56 compared to a state-of-the-artprior work Moons and Verhelst generalized the DynamicVoltage Accuracy Scaling concept to pipelined structuresand quantified its energy overhead in [13] DVAS technologyincludes three parts energy saving through voltage scalingaccuracy scaling through bit width reduction and voltagescaling through bit width reduction DVAS is finally appliedto a JPEG image processing application demonstrating largesystem level gains Arroba et al proposed a DVFS policy thatreduces power consumption while preventing performancedegradation and a DVFS-aware consolidation policy thatoptimizes consumption considering the DVFS configurationthat would be necessary when mapping virtual machines tomaintain Quality of Service in [14] Han et al proposed anoff-line dynamic voltage scaling (DVS) scheme that can beintegrated with EDF which is a global real-time schedulingalgorithm for symmetric multiprocessor systems in [15]However these techniques are unsatisfactory in minimizingboth schedule length and energy consumption

Resource hibernation is the setting or prediction of theclosingsleeping time for computer processing componentsThere are many challenges for resource hibernation technol-ogy in cloud computing systems with many cloud computingresources For example the shortcomings of the traditionalscheduling strategies can lead to computer load imbalancebesides using resource hibernation technologywill obviouslyseriously affect the performance of the entire system

However the above researches discuss energy-savingmethods only from the systemhardware management theappropriate task scheduling strategies can also achieve thegoal of saving energy in cloud data centers




3 System Model







Load (119894) = radic 1119899 119899sum119894=1














VM1VM2

VM1VM2

VM1VM2

VMa

VMb

VMn




Assigning tasks

DVS control


MaxVM(i))

Available VMs



Queue status

M1

M2

Mm








SPN = (119875 119879 1198651198821198720 120582) (4)





51 Prediction









p t

pa1

pa2

pa3

pam

ta1

ta2

ta3

tam

pb1

pb2

pb3

pbm

tb11

tb1n

tb2n

tb21

tb31

tb3n

tbm1

tbmn

pc1 tc1

pcn tcn

pd1

pdn

td pe




Time119895 (119905) = 1205720 + 1205721119880cpu119895 + 1205722119865cpu119895 + 1205723119872119895 + 1205724119878119895+ 120576 (5)













(8)




















Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of







3 System Model







Load (119894) = radic 1119899 119899sum119894=1














VM1VM2

VM1VM2

VM1VM2

VMa

VMb

VMn




Assigning tasks

DVS control


MaxVM(i))

Available VMs



Queue status

M1

M2

Mm








SPN = (119875 119879 1198651198821198720 120582) (4)





51 Prediction









p t

pa1

pa2

pa3

pam

ta1

ta2

ta3

tam

pb1

pb2

pb3

pbm

tb11

tb1n

tb2n

tb21

tb31

tb3n

tbm1

tbmn

pc1 tc1

pcn tcn

pd1

pdn

td pe




Time119895 (119905) = 1205720 + 1205721119880cpu119895 + 1205722119865cpu119895 + 1205723119872119895 + 1205724119878119895+ 120576 (5)













(8)




















Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





VM1VM2

VM1VM2

VM1VM2

VMa

VMb

VMn




Assigning tasks

DVS control


MaxVM(i))

Available VMs



Queue status

M1

M2

Mm








SPN = (119875 119879 1198651198821198720 120582) (4)





51 Prediction









p t

pa1

pa2

pa3

pam

ta1

ta2

ta3

tam

pb1

pb2

pb3

pbm

tb11

tb1n

tb2n

tb21

tb31

tb3n

tbm1

tbmn

pc1 tc1

pcn tcn

pd1

pdn

td pe




Time119895 (119905) = 1205720 + 1205721119880cpu119895 + 1205722119865cpu119895 + 1205723119872119895 + 1205724119878119895+ 120576 (5)













(8)




















Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





p t

pa1

pa2

pa3

pam

ta1

ta2

ta3

tam

pb1

pb2

pb3

pbm

tb11

tb1n

tb2n

tb21

tb31

tb3n

tbm1

tbmn

pc1 tc1

pcn tcn

pd1

pdn

td pe




Time119895 (119905) = 1205720 + 1205721119880cpu119895 + 1205722119865cpu119895 + 1205723119872119895 + 1205724119878119895+ 120576 (5)













(8)




















Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of









(8)




















Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





Start






Selecting next hop





Initialing ants

No


Yes



Yes


End

NoYes

yes

No

No












LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of














LET-ACOACO

Min-MinRR

0

500

1000

1500

2000

2500

Tota

l run

ning

tim

es (s

)

150 200 250 300 350 400100Number of tasks






LET-ACOACO

Min-MinRR

0

1000

2000

3000

4000

5000

6000

7000

8000

Ener

gy co

nsum

ptio

n (J

)

150 200 250 300 350 400100Number of tasks


100 150 200 250 300 350 400Number of tasks

LET-ACOMin-MinRR

0

5

10

15

20

25

Task

wai

ting

time (

s)





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of





LET-ACOMin-MinRR

0

01

02

03

04

05

06

07

08

100 200 300 400

Hos

t loa

d

Number of tasks

Figure 7 Host load





7 Conclusion






Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of







































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of











Volume 2014




Journal of











Journal of


Function Spaces






Algebra





Journal of




Documents

Ant Colony Optimization Algorithm to Dynamic Energy Management in Cloud …downloads.hindawi.com/journals/mpe/2017/4810514.pdf · Optimization of Energy Consumption in Cloud Data