Study of Heuristic MAP Selection and Abstraction Schemes with Load Balance in HMIPv6

Wireless Pers Commun (2011) 57:217–232DOI 10.1007/s11277-009-9854-5

Study of Heuristic MAP Selection and AbstractionSchemes with Load Balance in HMIPv6

Yen-Wen Chen · Ming-Jen Huang

Published online: 22 October 2009© Springer Science+Business Media, LLC. 2009

Abstract In order to improve the performance of real time services in the mobileenvironment, the hierarchical approach was proposed to reduce the frequency of handoffs inthe mobile IPv6 environment. The mobility anchor point (MAP) is adopted in a hierarchicalmanner to handle the location management of the mobile node within the MAP domain. Anenhanced speed estimation scheme is provided in this paper to select the appropriate MAPso that the overhead of handoff can be minimized. Furthermore, the concept of abstract MAP(AMAP) node is proposed to extend the coverage domain of MAP. In addition to selecting aMAP with respect to the speed estimation of the mobile node in a heuristic way, the criteriaof load balance is also investigated in this study. The performance of the proposed scheme isexamined through exhaustive simulations. And the simulation results show that the proposedscheme can effectively achieve the above objectives.

Keywords Hierarchical mobile IPv6 · Mobility anchor point · Handoff · Binding update

1 Introduction

The integration of wireless technology and Internet has made the Internet services very acces-sible. Furthermore, various kinds of services and applications being provided on the Internetlead to a huge increase in both the numbers of Internet users and information providers. Asthe number of devices connected to Internet increases, the demand for more IP addressesbecomes urgent accordingly. Although several approaches, such as network address transla-tor (NAT), were developed to solve the problem of insufficient IP address, some limitationswere introduced during service deployment. The IP protocol version 6 (IPv6) is treated as themost convincing protocol for the coming next generation all-IP network because of its largeaddressing space. In order to provide constantly connected service, technology of mobilitymanagement, which provides seamless handoff, has become one of the most important issues

Y.-W. Chen (B) · M.-J. HuangDepartment of Communication Engineering, National Central University, Jhongli, Taiwan, ROCe-mail: [email protected]

123

218 Y.-W. Chen, M.-J. Huang

when deploying wireless services. Mobile IP was proposed by the organization of InternetEngineering Task Force (IETF) as the mechanism of layer 3 handoff to support mobile deviceswith the transparent routing of IP packets during its movement [1,2]. The key idea of mobileIP is to assign the Home Agent (HA) of the Mobile Node (MN) as a registration point forthe Care-Of-Address (CoA) of the mobile node when it visits a foreign network. Any Corre-spondent Node (CN) can send packets to the mobile node through HA by using the tunnelingpath from HA to MN. Furthermore, the Route Optimization (RO) approach was proposedfor the direct path between the correspondent node and the mobile node [2,3].

Compared with IPv4, the newly deployed version 6 protocol allows greater overall consid-erations for the provisioning of mobility. For example, the route optimization is determinedas a fundamental part of the mobile IPv6; the security protocol (IPSec) is applied for theprotection of binding update messages among the mobile node, correspondent node, andhome agent; the auto-configuration mechanism makes the obtaining of the care of addresseasier; the extension headers (e.g. home address option and type 2 routing header) providea suitable way to deal with the problem that may occur in some IP services; etc.. Althoughthe route optimization can provide the direct transmission path from correspondent node tothe mobile node, the process of binding update is complex. For example, the procedure ofreturn routability is necessary for security considerations. This procedure shall be performedwhenever the mobile node moves into a new foreign network. Therefore, even by using ROprocedure, the quality of service may also be affected if a mobile node is moving at a highspeed and too many handoffs take place.

The performance caused by handoff in the mobile environment can be improved throughtwo approaches. One is to reduce the handoff latency and the other is to enlarge the routingdomain so that the frequency of handoffs can be minimized. The fast handoff procedure [4]was proposed to reduce handoff latency by processing the configuration of CoA and Dupli-cated Address Detection (DAD) in advance. The hierarchical mobile IPv6 (HMIPv6) [5]was proposed to reduce the number of remote binding updates during the movement of themobile node by selecting a mobility anchor point (MAP) at the suitable hierarchical layer.The efficiency of the fast handoff procedure depends on that the Access Router (AR) ofthe mobile node, which is going to visit, shall be predicable (e.g. train, car in the freeway).However, if the mobile node moves randomly (e.g. car/people moving around in the city),it is not easy to predict the next AR. Thus, the fast handoff scheme is suitable to be appliedfor MN with stable movement. The main idea of HMIPv6 is to select a suitable MAP so thatthe mobile node does not need to perform binding updates with its home agent. In this paper,we focus on the issue of the selection of the MAP node. MN has no need to inform its homeagent when traveling within the MAP domain. Generally, a higher layer MAP, if its load issustainable, has larger management domain and is more appropriate to be assigned to thenodes with higher mobility. Both of the moving speed of the mobile node and the load ofMAP are the most important criteria for the selection of MAP in HMIPv6. Recently, severalstudies [10–13] have been devoted to this issue. Most of them select the MAP for a mobilenode by estimating the moving speed of the mobile node. However, the estimated movingspeed may greatly differ to the actual speed because the speed is estimated only from thedwelling time at the previous AR and the coverage of that AR. Furthermore, a mobile nodewith low speed may also have a very short dwelling time in an AR with wide coverage.For example, if a low speed mobile node only crosses over a corner of the previous AR.In this paper, we suggest that the speed of the mobile node be estimated from the corre-lation of the latest estimated speed and average speed so that the predicted dwelling timecan be more accurate. And the weight factor for the above two speeds is determined by thevariation between them. Moreover, the concept of abstract MAP node (AMAP) in HMIPv6

123

Study of Heuristic MAP Selection and Abstraction Schemes 219

architecture is proposed so that the management domain of MAP can be enlarged and thefrequency of remote binding updates can be further reduced.

This paper is organized as follows. In the following section, the operation scenario MAPand related studies are briefly reviewed. The proposed speed estimation and the AMAP oper-ational scenario are stated in Sect. 3. In Sect. 4, we examine the performance, includingthe latency of binding update and the load distribution, of the proposed scheme throughexhaustive simulations. Finally, conclusions are provided in the last section.

2 Overview of Related Works

As mentioned in previous section, too frequent binding updates of mobile node will reducethe quality of services of real time applications. The basic idea of HMIPv6 is to deal withthe binding update procedure in a hierarchical way. A router is chosen by the mobile nodeas an anchor point (or local home agent), i.e. MAP, to limit the amount of MIPv6 signalingoutside the local domain. Two kinds of CoA are defined in HMIPv6. The Regional Care ofAddress (RCoA) can be treated as the address supported by MAP for the registration andbinding update at the HA and CN of the mobile node. The on-Link Care of Address (LCoA)is configured by MN based on the prefix advertised by current AR. LCoA is also used as anidentification of MN when traveling within the MAP coverage domain. In accordance withthe IETF specification [5], the Router Advertisement (RA) with MAP option is periodicallysent from the highest AR to the lowest AR to inform the mobile nodes of the status, includingvalidity, preference, distance, etc. for each MAP. And the mobile node can select a suitableMAP for binding update and configuring its RCoA when it moves into a new MAP domain.Further, the remote binding update procedure, which is the same as the procedure performedin original mobile IPv6, shall be performed by the mobile node to inform its HA and CNof its current RCoA. The mobile node will not change its RCoA if the movement of MNis within the coverage of the selected MAP and it is not necessary to inform its HA andCN. Thus, when MN moves into a new AR of the same MAP domain and obtains a newLCoA, it only needs to inform its MAP for a local binding update (LBU). The latency of localbinding update is much less than that of remote binding update. The operational procedureof HMIPv6 is illustrated in Fig. 1. MAP shall maintain a mapping table for each pair ofRCoA and LCoA for packet encapsulation. Packets sent to MN will firstly be routed to MAPaccording to RCoA because of the remote binding update, then tunneled by MAP with LCoAof MN and forwarded to MN.

As the network is hierarchically constructed in HMIPv6, it is easy to deduce that to selecta higher layer MAP for a mobile node means the mobile node associated with it will have awider coverage domain MAP. The existence of a wider coverage domain indicates that thefrequency of remote binding updates performed by this mobile node can be reduced becausethe RCoA of the mobile node is unchanged. However, there are some concerns associatedwith the selection of a higher layer MAP. One is that a higher layer MAP will introducelonger latency in the local binding update when compared to a lower layer MAP. The otheris that, in accordance with the characteristics of hierarchical topology, the number of ARdecreases as the hierarchical layer is getting higher. It is not appropriate to assign the highestMAP for all (or most) mobile nodes within that MAP domain because of the consideration ofthe processing load. Because MAP shall execute the translation of RCoA and LCoA, packetencapsulation for tunneling, etc., and the delivery of QoS packets will be a problem if theMAP is overloaded. Therefore, the load among MAP nodes shall be properly distributed. Areasonable approach is to assign higher layer MAP to the mobile nodes with higher moving

123


Inter-domain Handoff

HA

CN

Internet

New MAP

AR1

MN

New MAP Domain

LCoA RCoA

2. BU to HA or CN

RCoA HoA

RCoA HoA

MNMNIntra-domain Handoff

Old MAP

Old MAP Domain

1.LBU to new MAP and old MAP

Fig. 1 Operational procedure of HMIPv6

speed and to select the lower layer MAP for the mobile nodes with slower speed. Followingthis concept, to predict or estimate the moving speed of the mobile node becomes an essentialpart of this issue. In [11–13], several speed prediction schemes were proposed based on thespeed estimated in previous traveling domains. For example, in [12], the moving speed is cal-culated from the average of the dwelling time of the mobile node in the previous AR and thestandard coverage of an AR. The speed of the mobile node is estimated by the autocorrelationof the calculated speed of the mobile node in previous AR and the speed estimated when themobile node moved into the previous AR. Thus, assume that the mobile node travels fromAR(1), AR(2),…, AR(i), … and let SC(i) and SE(i) be the calculated speed in AR(i) and theestimated speed when the mobile node was just moving into AR(i), respectively. Thus, SE(i)

is the speed estimated for the reference in AR(i + 1). And if the mobile node is going to leavethe current MAP domain, then SE(i) is adopted for the selection of a suitable MAP. In [13],SC(i) and SE(i) were calculated as

SC(i) = d

Tdwell(i)(1)

SE(i) = αSC(i) + (1 − α)SE(i−1) (2)

where d is the standard coverage of an AR, Tdwell(i) is the dwelling time of the mobile nodein AR(i), and the value of αt ′is set between 0 and 1. The speed estimated in AR(i), SE(i), isapplied by the mobile node to select the suitable MAP by referring to the Selection Table(ST) advertised by MAP nodes.

123


3 Statistical Speed Estimation and Abstract MAP

Although the correlation property of the mobile speed was considered in [12,13] to increasethe accuracy of the prediction, it may also cause large estimation errors especially when thetravel distance of the mobile node in AR is much shorter than the standard distance d . Forexample, as shown in Fig. 2, the dwelling time of a mobile node in AR2 is noticeably shorterthan that of AR1 and AR3. When the mobile node is going to leave AR2 to enter AR3, itsspeed that will be in AR3, SE(2), can be predicted by using SC(2) and SE(1). If the weight forthe reference of calculated speed in AR2 (SC(2)), α, is very high (i.e. very close to 1), then thespeed estimated in AR3 will be much higher than the actual speed due to the short dwellingtime in AR2.

In this paper, a novel speed prediction scheme based on the average dwelling time andits standard deviation from the mobile node is proposed to improve this situation. And thevalue of parameter α is also determined in a systematic and heuristic way. Furthermore, theconcept of abstract MAP node is proposed so that the total number of remote binding updatesand the handoff latency can be reduced.

3.1 Speed Estimation of Mobile Node

In the proposed scheme, the speed of the MN is estimated when it is going to change itsMAP domain. Within the same domain, the mobile node accumulates the dwelling time andnumber of AR it has traveled. In [12], the dwelling time is considered as the time that MNstays in the previous AR. However, in HMIPv6, a mobile node may travel a lot of ARswithout changing its RCoA prior to leaving current MAP domain. Let n be the number ofARs the mobile node has visited and Tacc be the accumulated dwelling time of the visitedAR. We can easily calculate the average dwelling time in an AR by dividing Tacc by n. And,basically, the moving speed of the mobile node has a temporal correlation. Therefore, wedefine a threshold value m to specify the maximum number of the latest visited AR whosedwelling records will be taken into accumulation. Let Tμ be the estimated average dwellingtime for the mobile node in an AR, and we have,

Tμ = Tacc

Max(m, n)(3)

Fig. 2 Difference of dwellingtime in AR

123


By using autocorrelation, we can estimate the dwelling time of the mobile node inAR(i), T ′

dwell(i), as

T ′dwell(i) = (1 − α)T ′

dwell(i−1) + αTμ (4)

and the speed of the mobile node can be calculated as

S′E(i) = d

T ′dwell(i)

(5)

The value of α is treated as the parameter for adjusting the weight of SC(i) and SE(i) as statedin Eq. 2. The selection of its value was not specified clearly in [12]. In this paper, the valueof α is determined by the degree of correlation as follows:

α ={

1 if |T ′dwell(i−1) − Tμ| ≥ Tσ

|T ′dwell(i−1)

−Tμ|Tσ

otherwise(6)

where Tσ is the standard deviation of the dwelling time for the mobile node in an AR. Themain consideration of Eq. 6 is that if the difference between the dwelling time in the previousAR and the average dwelling time is greater than the standard deviation, then the dwellingtime in previous AR will be ignored and the average dwelling time will be used as the esti-mated dwelling time. On the other hand, if the difference is less than the standard deviation,the value of α is set as linearly proportional to the degree of difference. Thus, Tσ is regardedas a threshold for the consideration of referencing weights.

In order to illustrate the efficiency of our estimation scheme, the following example isprovided. Consider a mobile node travels in a topology at a constant speed of 10 m/s as shownin Fig. 3.

The moving speeds estimated in each AR by using Eqs. 1 and 2 and the proposed schemeare listed in Table 1 for the purpose of comparison. For example, the estimated speed atAR-2 when MN is going to move to AR-2 is 12.5 m/s (500/40) and the calculated speed atAR-2 is 10 m/s (500/50), then the estimated speed at AR-3 is 11.25 m/s (10 × 0.5 + 12.5 ×

Fig. 3 Example of the movement of a mobile node

123


Table 1 Speed estimation ofeach AR

AR Dwell time(s)

Speed estimatedby Eqs. 1 and 2

Speed estimatedby the proposedscheme

AR-1 40 – –

AR-2 50 12.5 12.5

AR-3 50 11.25 10.76

AR-4 35 10.63 10.4

AR-5 55 12.46 11.43

AR-6 30 10.77 10.87

AR-7 20 13.72 11.54

AR-8 30 19.36 12.5

AR-9 40 18.01 13.87

AR-10 50 15.26 12.53

AR-11 – 12.63 12.0

(1 − 0.5)) by using Eq. 2 and assuming α equals to 0.5. For the same estimation of theproposed scheme, the values of Tμ and Tσ calculated from AR-1 and AR-2 are 45 and 7.071,respectively and α equals to 0.707, i.e. (50−45)/7.071, by using Eq. 6. The estimated speedat AR-3 of the proposed scheme is, therefore, equals to 10.76 m/s by using Eqs. 4 and 5.It is noted that the speeds estimated by Eqs. 1 and 2 are much larger than the actual speed(10 m/s) in AR-8, AR-9, and AR-10 (the differences are 9.36, 8.01, and 5.26 m, respectively).However, the largest estimation error of our scheme is only 3.78m (in AR-9), which is muchsmaller than the error estimated by using Eq. 1 and 2. The main reason is that the schemeof Eqs. 1 and 2 may propagate the estimation error while our scheme can terminate theestimation error by considering the standard deviation as indicated in Eq. 6.

3.2 Abstract MAP

In HMIPv6, each MAP has its management domain and the mobile node can have the advan-tage of saving on the overhead of remote binding update when it moves inside a MAP domain.Therefore, in order to reduce this overhead, selecting a higher layer MAP can overcome thisproblem. However, as mentioned in Sect. 2, the number of MAP nodes decreases as the layergets higher and the load balance and path used for local binding update are also the maincriteria that affect the performance in handoff procedure.

In order to specify the management capability of MAP in HMIPv6 architecture, an index,denoting the degree of MAP coverage as D, is defined in this paper to represent the capabilityin managing handoff of a specific network in HMIPv6. Assume that a network with n-layerMAP is partitioned into several regions and each region is served by an AR. Let di be themanagement capacity of the layer-i MAP and is defined as

di = di,c×ri (7)

where di,c denotes the number of mobile nodes of a MAP of the layer-i can handle. ri is thenumber of regions covered by each MAP of the layer-i . Then, the degree of MAP coverageD is defined as

D =∑n

i=1 di

R∑n

i=1 di,c(8)

123


Gateway MAP

Layer3 MAP1 Layer3 MAP2

L2 MAP1 L2 MAP2 L2 MAP3 L2 MAP4

L1 MAP1 L1 MAP2 L1 MAP3 L1 MAP4 L1 MAP5 L1 MAP6 L1 MAP7 L1 MAP8

Abstract MAP nodeMN

AR1 AR2 AR3 AR4 AR5 AR7 AR8AR6 AR9 AR11 AR12AR10 AR13 AR15 AR16AR14

Fig. 4 Abstract MAP nodes in hierarchical IPv6

where R is the total number of regions, i.e. R = Max(ri ). A larger value of D indicates thatthe network has larger capacity and better performance in dealing with handoff. If D equals to1, it means that each mobile node can travel in all regions without any remote binding updateno matter which MAP it selects. For example, consider a network with 16 regions (AR) and4-layer MAP as shown in Fig. 4, the numbers of regions that can be handled by each MAPfrom the lowest layer to the highest gateway MAP are 2, 4, 8, and 16, respectively. Assumethe numbers of mobile nodes that the MAP can handle from the lowest layer MAP to thehighest gateway MAP are 20, 30, 50, and 80, respectively. Then, the management capabilityof each MAP located at layer 1 is d1 = 20 × 2 = 40. And the values of d2, d3, and d4 are120, 400, and 1280, respectively. As the numbers of MAP nodes of each layer are 8, 4, 2,and 1, the proposed degree of MAP coverage D can be obtained as

D = 40 × 8 + 120 × 4 + 400 × 2 + 1280 × 1

16 × (20 × 8 + 30 × 4 + 50 × 2 + 80 × 1)≈ 0.39

In this paper, the concept of abstract node is introduced to logically extend the coverage ofMAP. Generally, an abstract MAP, AMAP, is a group of MAP nodes that cooperate togetherfor a wider management domain. Here we assume that all MAPs are managed by one serviceprovider or cooperative providers and the security issue is not discussed in this paper. Asshown in Fig. 4, three MAP nodes located at layer 1 and layer 2 of a hierarchical topology,are grouped together as an AMAP. Then, the service domain of the abstract node is the sameas the MAP located at layer 2. However, these three MAP nodes can serve this domain in aload balanced manner. In this case, the number of regions managed by original L1 MAP ischanged from 2 to 4. And the degree of MAP coverage D increases from 0.39 to 0.44 becaused1 becomes 80.

It is also noted that if a mobile node is served by a MAP, whose original managementdomain does not cover the current location of the mobile node of an abstract MAP node,the forwarding path of the packets will travel backward to this MAP and forward to thecurrent MAP and AR where the mobile node originally resided. This backward and forward

123


MAP-1

MAP-2

MAP-3

MAP-2 Domain

MAP-3 Domain

Abstract MAP

MN

MN

LBU

Internet

Data Packet

LBAck

Fig. 5 Local binding update and packet forwarding of AMAP

movement may cause extra path overheads in transmitting packets as shown in Fig. 5. It showsthat the mobile node does not need to change MAP when it moves from the MAP-2 domainto the MAP-3 domain because MAP-1, MAP-2, and MAP-3 are grouped as an AMAP. TheLBU message and the tunneling of data packets will be processed by MAP-2. Although extraoverhead exists for the round trip communication between MAP1 and MAP2, this overheadcan be greatly reduced if all MAPs of the same abstract node are connected with each otherby asynchronous transfer mode (ATM) connections or label switch path (LSP).

In order to construct AMAP and to inform mobile nodes which MAP nodes are groupedas an abstract node, two fields in the reserved part of the MAP option message in neighbordiscover extension of the IETF RFC [5] are added. A one bit field, A, is used as the indicationto ask its successors for grouping and the field of S is used for scoping the size of an AMAP.The size of the AMAP is defined as the number of hop counts from the highest layer MAPof this AMAP to the lowest ones.

The construction procedure of an AMAP is explained through the following example. Thelayer 2 MAP in Fig. 4 is designed as the initiating node for the construction of an abstractnode. The initiating MAP node shall determine the size (scope) of the abstract node in thebeginning. The MAP option is then sent with RA by setting the A bit (A = 1) and the desiredsize, S = 1, to its downstream nodes (layer 1 MAP). Upon receiving the message of the RA,the layer 1 MAP nodes shall examine the fields of A and S. If A is set and the field size isgreater than zero, the MAP shall decrease the field size by 1, attach its MAP option with thisRA, and forward the attached RA message to its successors. If the value of the size equalsto zero, it means that the receiving MAP does not belong to the abstract node. In order toprovide information of all MAP nodes to all mobile nodes within the domain of the sameabstract node, a backward RA is also sent from the lower layer MAP to its precedence nodeso that the highest layer MAP of the abstract node can have complete information availablefor distribution. The scenario of the construction of the abstract is shown in Fig. 7. Thedestructuralization of an abstract node can also be achieved in a similar way to the AMAPconstruction. The procedure is also initiated by the highest layer MAP through RA messageas before, however, the A bit is set as zero for the purpose of destructuralization.

After receiving the RA message, the mobile node obtains the MAP selection table listingall available MAP nodes (including the AMAP nodes). All MAP nodes belonging to thesame AMAP have the same management domain. The mobile node can then use Eq. 5 toestimate its speed and to select a suitable MAP node when it is going to move into a new

123


Fig. 6 Modified MAP optionmessage

Type Length Dist. Pref. R A S

Valid lifetime

Global IP Address

for MAP

MAP-B

AR1 AR2

MN

AR3 AR4

MAP-A

RA RARA RA

RAMAP Option

A flag is set ; Size = 1

RA MAP-C

RA RA

MAP OptionA flag is set ; Size = 1

MAP OptionA flag is set ; Size = 0

Abstract MAP Node

Fig. 7 Operational scenario of abstract node construction

MAP domain. It is noted that the sustainable load of a MAP shall also be considered duringthe selection of MAP. Assume that the sustainable number of mobile nodes to be managedby a MAP is MAPMN. Let MAPL be the number of mobile nodes currently being served bythe MAP. Then, the preference field of the MAP option message can be set as

preference = 15 ×(

1 − MAPL

MAPMN

)(9)

The maximum preference value is 15 and the larger the preference value is the lighter theload of the MAP is. And, for all MAP nodes of an AMAP, the mobile node shall select theMAP with the larger preference value in its selection table for load balance.

4 Experimental Simulations

The performance of the proposed scheme is examined through computer simulations. Thesimulation program is written in C. The network topology for simulation is of three-layers asshown in Fig. 8. Each lowest MAP accommodates 4 ARs. The delay times of links betweenconsecutive layers are 2, 4, and 6 units, respectively. The delay time from the highest layerMAP to CN is assumed to be 100 time units. The coverage of each AR is 500 m × 500 m and,

123


Fig. 8 Simulation topology

Table 2 Moving behaviors ofhigh mobility and low mobilityMN

MN speed Dwelling time

High mobility MN 15−30 m/s (54−108 km/h) (0.9, 0 s)(0.1, 60 s)

Low mobility MN 1−15 m/s (3.6−54 km/h) (0.5, 360 s)

(0.3, 600 s)

(0.1, 120 s)

hence, the total coverage of the topology is 4000 m × 4000 m. The capacity (i.e. the numberof mobile nodes can be handled) of layer-3 MAP is 200 and it is 50 for each MAP of layer-2and layer 1.

The random waypoint mobility model [14] is adopted as the moving behavior of eachmobile node. Both high mobility and low mobility MN coexist in our simulations. The rateof high mobility MN is defined as the ratio of the number of nodes with high mobility to thetotal number of mobile nodes. Table 2 specifies the parameters assumed for the movementof high mobility and low mobility MN. The moving speed is uniformly distributed withinthe specified range and the dwelling time is within the probability-based hyper exponentialdistribution. For example, the term of (0.1, 60 s) means that there is 10% possibility of an MNstaying in its current location for a period when it moves into an AR while the probabilitythat MN will keep moving is 90%. This period of stay is hyper exponentially distributed witha mean of 60 s.

Three schemes are applied in our simulations for the purpose of comparison. The first oneis the traditional farthest MAP first scheme, in which the mobile node chooses the highestMAP if its capacity is affordable. The second one is the mobility based scheme, which usesthe speed estimation scheme we proposed in this paper as the reference to select a suitableMAP. For the third one, in addition to using the above speed estimation, the concept of theabstract MAP node is also applied. In the scheme of abstract MAP node, two kinds of node

123


abstraction are adopted. The high layer abstraction is to group layer 2 and layer 3 MAP nodesinto an abstract MAP node; and the low layer abstraction is to group layer 1 and layer 2 MAPnodes into abstract MAP nodes. The degree of MAP coverage D for the topologies withoutabstraction and with abstraction (either high layer or low layer) can be calculated, accordingto Eqs. 7 and 8, as 19200/76800 = 0.25 and 28800/76800 = 0.375, respectively. The valueof D increases because of the affect of MAP abstraction.

Figures 9 and 10 illustrate the number of inter-domain handoffs (i.e. remote bindingupdates) for the low mobility and high mobility environment, respectively. In Fig. 9, it showsthat there is no inter domain handoff when the number of mobile nodes is less than 200 (thesustainable load of the highest MAP) for tradition farthest MAP first scheme. Because, in thiscase, all mobile nodes can select the highest MAP (i.e. layer-3 MAP) and no remote bindingupdate is necessary. However, the number of handoff dramatically increases when the num-ber of mobile nodes exceeds 200. The reason is that this scheme does not consider the speedof the mobile node in the selection of its MAP. Then, the layer-3 MAP may be selected asMAP by several low-speed mobile nodes and these low-speed MN will not change their MAPbefore the ending of their connections. If the load of layer-3 MAP is saturated, it will forceother MN, including high-speed and low-speed MN, to select a lower layer MAP. Hence, as

Fig. 9 Inter-domain handoff of20% high mobility rate

Fig. 10 Inter-domain handoff of80% high mobility rate

123


more high-speed mobile nodes select lower MAP, remote binding update tends to occur. Theproposed schemes, including solely mobility based and the abstract MAP, illustrate quite astable number of handoff even when the number of mobile nodes exceeds 500. It is notedthat the high layer abstraction has the least number of inter-domain handoffs because, dueto the abstraction, the layer-2 MAP will be “upgraded” to cooperate with the layer-3 MAPfor the routing management of the entire simulation network. And no inter-domain handoffis required for the mobile nodes to select either layer 2 or layer 3 MAP. In the high mobilityenvironment (Fig. 10), the proposed schemes also demonstrate superior performance to thatof the traditional schemes. It is also noted that the number of inter-domain handoffs starts toincrease when the number of MN is 450. The reason is that, in the high mobility environ-ment, there are about 360 mobile nodes (450 × 80% = 360) having high mobility behavior,which is approaching the sustainable load of the abstracted node. The high layer abstractnode includes one layer 3 MAP node and four layer 2 MAP nodes and its sustainable loadis 200 + 50 × 4 = 400. Thus, selecting the higher layer MAP can effectively reduce thenumber of handoffs. And, because the degrees of MAP coverage of high layer and low layerabstractions are the same, we found that the numbers of inter-domain handoffs are becomingclose when the number of MN exceeds 500.

The performance of binding update latency (in milliseconds) for the high mobility envi-ronment is shown in Fig. 11. The impact of the backward and forward phenomenon of bindingupdate messages occurred in AMAP shown in Fig. 5 is also taken into account during sim-ulations. We observe that the proposed scheme illustrates a lower latency than that of thefarthest first scheme. It is also noted that, when the number of mobile nodes exceeds 500,the binding update latency of the low layer abstract MAP scheme starts to become lowerthan that of the high layer abstract MAP scheme. The main reason of this phenomenon isthat the low layer abstraction groups more MAP to be AMAP and the overall load is sharedby more MAP nodes when compared to the high layer abstraction scheme. In the high layerabstraction case, when the loads of layer 2 and layer 3 MAP are saturated (i.e. the numberof MN is greater than 400), the layer 1 node will be selected as MAP and it will cause morehandoffs for high mobility nodes because there is no abstraction at the lower layer.

The results of load distributions of the MAP at each layer for the farthest, mobility based,and high layer abstraction schemes with high mobility environment are depicted in Figs. 12,

Fig. 11 Binding update latency

123


13, and 14, respectively. For the farthest first scheme, the condition of load saturation regu-larly occurred from the layer 3 down to layer 1. Further, the load of each layer starts increasingafter its upper layer is saturated (i.e. 200 and 400, respectively). In Fig. 13, the load of layer 3MAP is saturated when the number of MN is greater than 300 for the mobility based scheme.As shown in Fig. 14, the load of AMAP, which consists of layer 2 and layer 3 MAP, is effec-tively shared by the constituted MAP nodes. And their loads are saturated when the numberof mobile nodes exceeds 450, which is approaching the sustainable load of the abstractednode as explained in Fig. 10. This demonstrates that the proposed scheme has a better loadbalance than other schemes.

5 Conclusions

In this paper, the speed estimation is improved by considering the correlation between thelatest estimated speed and the average speed and the affect of the standard deviation. The

Fig. 12 Load distributions of farthest MAP scheme

Fig. 13 Load distributions of mobility based scheme

123


Fig. 14 Load distributions of abstract MAP scheme

experimental results indicate that the proposed estimation scheme is more accurate whencomparing to the scheme proposed in [13]. Furthermore, the concept of an abstract MAPnode is proposed in this paper to decrease the frequency of handoff and to achieve loadbalance in the HMIPv6 environment. The modifications of the MAP option message in theneighbor discover extension is stated in order to deal with the construction and destructural-ization of an abstract MAP node. The degree of MAP coverage is firstly defined to representthe capability of handoff management. This degree value will be very helpful for the mea-surement of handoff quality in mobile environment. Exhaustive simulations were performedto examine the performance of the proposed scheme. The simulation results show that boththe performance of binding update and the load balance of MAP can be effectively improved.By applying the concept of abstract node, the construction and the destructuralization of anabstract MAP node may be configured more flexible with respect to the mobile behavior ofMN and the load condition of MAP. This continues to be one of our on going directions forfuture research.

Acknowledgments This research work was supported in part by the grants from the National ScienceCouncil (NSC) (grant numbers: NSC96-2221-E-008-011, 97-2221-E-008-033, and 98-2221-E-008-063).

References

1. Perkins, C. (2002). IP mobility support for IPv4, IETF RFC-3344, Nokia Research Center.2. Johnson, D., & Perkins, C. (2004). Mobility support in IPv6, IETF RFC-3775, Nokia Research Center.3. Nikander, P., Arkko, J., Aura, T., Montenegro, G., & Nordmark, E. (2007). Enhanced route optimization

for mobile IPv6, IETF RFC 4866.4. Koodli, R. (2008). Fast handovers mobile IPv6, IETF RFC-5268.5. Soliman, H., Castelluccia, C., El Malki, K., & Bellier, L. (2005). Hierarchical mobile IPv6 mobility

management (HMIPv6), IETF RFC-4140.6. Kent, S., & Atkinson, R. (1998). IP encapsulating security payload (ESP), IETF RFC-2406.7. Kent, S., & Atkinson, R. (1998). IP authentication header (AH), IETF RFC-24028. Arkko, J., Devarapalli, V., & Dupont, F. (2004). Using IPsec to protect mobile IPv6 signaling between

mobiles and home agents, IETF RFC-3776.9. Narten, T., Nordmark, E., & Simpson, W. (1998). Neighbor discovery for IP version 6 (IPv6), IETF

RFC-2461.

123


10. Yi, X. H., Lee, C. J., & Thing, V. L. L. (2003). A local mobility agent selection algorithm for mobilenetworks. IEEE International Conference on Communications (ICC), 2, 1074–1079.

11. Kawano, K., Kinoshita, K., & Murakami, K. (2002). A mobility-based terminal management in IPv6networks. IEICE Transactions Communications, E85-B(10), 2090–2099.

12. Kawano, K., Kinoshita, K., & Murakami, K. (2004a). A study on estimation of mobility of terminalsfor hierarchical mobility management scheme. IEICE Transactions Communications, E87-B(9).

13. Kawano, K., Kinoshita, K., & Murakami, K. (2004b). Multilevel hierarchical mobility managementscheme in complicated structured networks. IEEE International Conference on Local ComputerNetworks(LCN), 34–41.

14. Bettstetter, C., Hartenstein, H., P’erez-Costa, X. (2002). Stochastic properties of the random waypointmobility model, ACM Mobicom Workshop MSWIM.

Author Biographies

Yen-Wen Chen received the Ph.D. degree in electronic engineeringfrom National Taiwan University of Science and Technology (NTUST)in 1997. Since 1983, he jointed the department of switching technol-ogy, Chunghua Telecommunication Laboratories, Taiwan, ROC, as aresearcher. From 1995 to 1998, he was a project manager of broadbandswitch systems to develop the permanent virtual connection broad-band ATM switching systems (VPX). From August 1998 to July 2000,he jointed the department of information management, Central PoliceUniversity. Since August 2000, Dr. Chen joints the department ofcommunication engineering as an assistant professor. Currently, he isan associate professor. During the academic period, he has authored orco-authored 3 books and has published about 60 professional researchpapers. Dr. Chen serves as the editor of Journal of Information Tech-nology and Applications (JITA) and Journal of Information, Technol-ogy and Society (JITAS). His current research interests include WiMax,mobility management, IP over SDH/DWDM, GMPLS, IPv6, and QoSmanagement.

Ming-Jen Huang received the M.S. degree of communication engineering from National Central Univer-sity, Taiwan, ROC in 2006. He works at GemTek corp. as a research engineer, Taiwan now. His researchinterests include VoIP. mobile network, and communication software.

123

Documents

Study of Heuristic MAP Selection and Abstraction Schemes with Load Balance in HMIPv6