DMAP: integrated mobility and service management in mobile IPv6

Wireless Pers Commun (2007) 43:711–723DOI 10.1007/s11277-007-9276-1

DMAP: integrated mobility and service managementin mobile IPv6 systems

Ing-Ray Chen · Weiping He · Baoshan Gu

Received: 14 August 2006 / Accepted: 30 January 2007 / Published online: 20 April 2007© Springer Science+Business Media B.V. 2007

Abstract Mobile IPv6 (MIPv6) is a work in pro-gress IETF standard for enabling mobility in IPv6networks and is expected to have wide deploy-ment. We investigate an integrated mobility andservice management scheme based on MIPv6 withthe goal to minimize the overall network signal-ing cost in MIPv6 systems for serving mobilityand service management related operations. Ourdesign extends IETF work-in-progress Hierarchi-cal Mobile IPv6 (HMIPv6) with the notion of dy-namic mobility anchor points (DMAPs) for eachmobile node (MN) instead of static ones for allMNs. These DMAPs are access routers chosen byindividual MNs to act as a regional router to reducethe signaling overhead for intra-regional move-ments. The DMAP domain size, i.e., the numberof subnets covered by a DMAP, is based on theMN’s mobility and service characteristics. Underour DMAP protocol, a MN interacts with its homeagent and application servers as in the MIPv6 pro-tocol, but optimally determines when and where

I-R. Chen (B) · W. He · B. GuDepartment of Computer Science, Virginia Tech, 7054Haycock Road, Northern Virginia Center, FallsChurch, VA 22043, USAe-mail: [email protected]

W. Hee-mail: [email protected]

B. Gue-mail: [email protected]

to launch a DMAP to minimize the network costin serving the user’s mobility and service manage-ment operations. We demonstrate that our DMAPprotocol for integrated mobility and service man-agement yields significantly improved performanceover basic MIPv6 and HMIPv6.

Keywords MIPv6 · Hierarchical MIPv6 ·Mobility management · Service management ·Service handoff · Dynamic mobility anchor point(DMAP) · Mobility handoff · Performanceanalysis · Petri net

1 Introduction

IPv6 networks have been deployed rapidly all overthe world. Mobile IPv6 (MIPv6) [5] is a networkprotocol for enabling mobility in IPv6 networks.It allows mobile nodes (MNs) to move within IP-based networks while maintaining on-going con-nections. MIPv6 has been flagged as the mobilitymanagement protocol for future all-IP mobile sys-tems and is expected to have wide deployment.With the anticipated increase in inexpensive, com-putationally powerful mobile devices runningmobile applications to access multimedia and dataservices over broadband wireless connections basedon IPv6, it is of fundamental importance to res-earch effective and scalable network management

712 I.-R. Chen et al.

schemes based on MIPv6 to reduce network costsin IPv6 networks.

In this paper, we investigate a scalable and effi-cient integrated mobility and service managementscheme, called DMAP, based on MIPv6 with thegoal to minimize the network cost in MIPv6 sys-tems for serving mobility and service managementrelated operations. Our design extends IETF work-in-progress Hierarchical Mobile IPv6 (HMIPv6)[10] with the notion of “dynamic” mobility anchorpoints (DMAPs) for each individual MN instead ofstatic ones for all MNs. These DMAPs are simplyaccess routers (ARs) chosen by individual MNsto act as a regional router to reduce the signal-ing overhead for intra-regional movements. TheDMAP domain size, or the number of subnets in aregion covered by a DMAP, is based on the mobil-ity and service characteristics of each MN.

We aim at the design of DMAP to integrateseamlessly with MIPv6. The goal is to identify theoptimal DMAP domain size dynamically to mini-mize the network communication overhead due tomobility and service operations induced by eachMN. When a MN enters a subnet, the MN deter-mines if the AR would be selected as the DMAP.If yes, the MN obtains a regional care of address(RCoA) from the DMAP and informs its homeagent (HA) and application servers of the RCoAaddress change. Essentially with DMAP, a MNinteracts with the HA and application servers fol-lowing the standard MIPv6 protocol but dynami-cally determines when and where to launch aDMAP so as to minimize the network cost.

Our proposed DMAP impacts the MIPv6 Inter-net standard and benefits IPv6 networks by effec-tively integrating mobility and service managementto reduce networking costs. DMAP determines theoptimal MAP domain size to minimize networkcosts for both mobility and service induced man-agement operations. We demonstrate the networkcost saving to be significantly better than that pro-vided by basic MIPv6 and HMIPv6 [10] which dealwith mobility management only. Furthermore, thenetwork cost reduction benefit as a result of apply-ing the proposed scheme would be cumulative andproportional to the number of MNs. In this paper,we will use the term “service area” interchangeablywith “MAP domain” since in our design we dealwith service management in addition to mobility

management. A location handoff refers to the MNmoving across a subnet boundary, while a servicehandoff refers to the MN moving across a DMAPdomain boundary.

The rest of the paper is organized as follows.Section 2 describes related work. Section 3 des-cribes our proposed DMAP scheme for integratedmobility and service management in MIPv6 envi-ronments. Section 4 develops a computational pro-cedure to determine the optimal service area forservice handoffs in order to minimize the networkcommunication cost induced by mobility and ser-vice management operations. In Section 5, we com-pare DMAP with basic MIPv6 and HMIPv6.Finally, Section 6 summaries the paper and out-lines some future research areas.

2 Related work

For next-generation mobile IPv6 networks, MNsare expected to be very active with significantmobility. The mobility rate with which subnets arecrossed by MNs can be high, causing a high signal-ing overhead for the MN to inform HA and CNsof the CoA address change. There have been ap-proaches [2,7,10,12,13] proposed to mitigate thishigh volume of network signaling cost, including,most noticeably, IETF work-in-progress MIP Reg-ional Registration (MIP-RR) [4], HierarchicalMIPv6 [10] and IDMP [7]. MIP-RR uses a Gate-way Foreign Agent (GFA) to provide a regionalCoA, which acts as a proxy for regional movementmanagement to keep track of the MN’s currentCoA as long as the MN moves within a region,thereby reducing the network signaling cost whenthe MN moves within a region. When the MNmoves to a new region, the regional CoA changeis informed to the HA. The design is for mobil-ity management only without considering servicemanagement-induced network cost.

Hierarchical MIPv6 (HMIPv6) [10] is designedto reduce the network signaling cost for mobilitymanagement based on the observation that statis-tically local mobility accounts for more than 60%of movements made by a MN. In addition to aCoA, a regional CoA (RCoA) is also allocatedto a MN whenever the MN enters a new MAPdomain. The HA and CNs ideally only know the

Integrated mobility and service management in MIPv6 713

MN’s RCoA, so whenever the MN moves across aMAP domain and triggers a RCoA address change,the new RCoA address needs to be propagated tothe HA and CNs. Whenever, a MN moves fromone subnet to another but is still within a regioncovered by a MAP domain, the CoA change isonly propagated to the MAP instead of to the HAand CNs, thus saving the signaling cost for mobilitymanagement. The number of subnets covered bya MAP domain is static in HMIPv6. This conceptcan be applied at multiple levels in a hierarchicalmanner [9,10] and can be combined with the useof Fast Handovers [2] that use forwarding pointersbetween the current and next subnets in a hybridmanner. MAPs in HMIPv6 are statically config-ured and shared by all MNs in the system. ARs areresponsible for announcing their MAP’s identityby means of router advertisement packets so thatroaming MNs would know if they have crossed aMAP domain and need to perform a RCoA updateto the HA and CNs. Our proposed DMAP extendsHMIPv6 by taking both mobility and service man-agement into consideration, such that each MNdiscriminatively selects ARs as MAPs and dynam-ically changes the size of a MAP domain (corre-sponding to a “service area”) based on its mobilityand service characteristics at runtime so as to min-imize the total cost due to the signaling cost asso-ciated with mobility management and the networkcommunication cost associated with service man-agement.

Another IETF work-in-progress draft IDMP [7]introduces the concept of domain mobility with adomain corresponding to a region in MIP-RR andHMIPv6, and a domain agent corresponding to aMAP in HMIPv6 to keep track of CoA of a MNas the MN roams within a domain. IDMP can becombined with fast handoff mechanisms utilizingmulticasting [6] to reduce handoff latency and pag-ing mechanisms to reduce the network signalingcost for intra-domain movements.

HMIPv6 and IDMP can effectively reduce net-work signaling cost for mobility management com-pared with basic MIPv6 and are scalable to a largenumber of MNs. However, there is no mechanismprovided to determine the size of a MAP domainin HMIPv6 (or a mobility domain in IDMP) for allMNs that would minimize the network cost, sincethey deal with mobility management only without

considering service characteristics of individualMNs. Our proposed DMAP scheme extendsHMIPv6 by considering both mobility and ser-vice management. It aims to determine the bestMAP domain size on a per MN basis in order toreduce the network signaling and communicationcost resulting from both mobility and service man-agement for future IP-based Internet applications.

3 DMAP for integrated mobility and servicemanagement

We propose DMAP that extends HMIPv6 for inte-grated mobility and service management to reducethe network signaling and communication over-head for servicing mobility and service inducedoperations. DMAP takes a step further beyondbasic MIPv6 and HMIPv6 which we consider asspecial cases designed for mobility managementonly. The essence of DMAP is the notion of inte-grated mobility and service management. This isachieved by determining an optimal service areasize corresponding to the size of a MAP domainin HMIPv6. This service area size is expressed interms of the number of subnets. It is dynamicallydetermined for each MN, taking into considerationof both mobility and service characteristics of theMN. We develop a computational procedure bywhich the optimal service area size can be deter-mined. The objective is to minimize the total net-work signaling and communication overhead inservicing the MN’s mobility and service manage-ment operations.

We implement DMAP by means of a “DMAPtable lookup” design described as follows, lever-aging binding request messages defined in MIPv6and HMIPv6. This design has the advantage of sim-plicity, scalability and efficiency and is HMIPv6-compliant. When a MN crosses a service area, itmakes the AR of the subnet just crossed as theDMAP as in HMIPv6. The MN also determinesthe size of the new service area (or MAP domain).Concurrently, it acquires a RCoA as well as a CoAfrom the current subnet and registers the addresspair (RCoA, CoA) to the current DMAP (the ARof the current subnet) in a binding request mes-sage. Note that the RCoA could be the same asthe CoA upon the MN’s entry into a new DMAP


domain. The MN also informs the HA and CNs ofthe new RCoA address change in another bindingmessage so that the HA and CNs would know theMN by its new RCoA address. When the HA andCNs subsequently send packets to the MN, theywould use the RCoA as the MN’s address.

A packet destined for RCoA will first be inter-cepted by the DMAP. By inspecting the addresspair (RCoA, CoA) stored in the internal table, theDMAP knows that the MN’s address is actuallythe CoA and will forward the packet to the MNthrough tunneling. If the RCoA and CoA are in thesame subnet, the DMAP can directly forward thepacket to the MN without using tunneling. Whenthe MN subsequently crosses a subnet but is stilllocated within the service area, it would inform theMAP of the CoA address change without inform-ing the HA and CNs to reduce the network signal-ing cost. This “DMAP table lookup” design mapsRCoA to CoA by having the current DMAP main-tain an internal table, so the DMAP can intercepta packet destined for RCoA and forward it to theMN’s CoA. It is efficient since the RCoA-CoArouting function can be performed efficiently byDMAPs (which are routers) through simple tablelookup operations. It is scalable because the designis scalable to a large number of MNs by havingall ARs in MIPv6 networks DMAP-enabled andrandomly spreading the routing and table lookupfunctions to all ARs in the network. In terms ofsecurity and fault tolerance, it can also leverage

existing solutions in HMIPv6 because this designis HMIPv6-compliant except that a MN dynami-cally selects ARs to be MAPs.

The concept of per-user service DMAP derivesfrom our earlier work on service management inwireless personal communication networks [3].The idea of determining a dynamic “service area”to minimize the network signaling cost is similarin concept to determining a gateway foreign agent(GFA) coverage area in MIP Regional Registra-tion [12]. However, MIP Regional Registrationonly deals with mobility management while ourscheme deals with both mobility and service man-agement.

A MN’s service area can be modeled as consist-ing of K IP subnets. We develop a computationalprocedure to determine the optimal service areasize in terms of K. The optimal K value would bedetermined at runtime dynamically to minimizethe network cost. For the special case in which Kis constant for all MNs, our scheme degenerates toHMIPv6 with a two-level MAP-AR structure.

Figure 1 illustrates DMAP in MIPv6 environ-ments. When moving from one subnet and anotherwithin service area 1, the MN only informs its CoAchange to the DMAP without informing the HAor CNs. When the MN exits service area 1, the ARof subnet B becomes the DMAP. The MN obtainsa CoA and a RCoA from subnet B and an entry(RCoA, CoA) is recorded in the routing table ofthe AR of subnet B. Subnet B’s AR now acts as

Fig. 1 DMAP: integratedmobility and servicemanagement in MIPv6

Home Agent Correspondent NodeIP Network

A B

C

D E

Service Area 1 Service Area 3

Service Area 2


the DMAP of the MN. Both the HA and CNs areinformed of the RCoA address change by the MN.

In our DMAP scheme, the MN appoints a newDMAP only when it crosses a service area whosesize is determined based on the mobility and ser-vice characteristics of the MN in the new servicearea. One should note that the service area sizeof the DMAP is not necessarily uniform. In theabove scenario although subnet B appears to beat the edge of service area 2, it is actually at thecenter of the new service area since our servicearea is defined by the number of subnets (or thenumber of moves) starting from the first subnetat which the MN enters into a new service area.Within service area 2 (the new service area thatthe MN just moves into), if the MN moves fromsubnet B to subnet D through C (with 2 locationhandoffs), then the DMAP will be informed of theCoA change by the MN but will remain at the samelocation (subnet B).

The optimal service area size is dictated by theMN’s mobility and service characteristics. A largeservice area size means that the DMAP will notchange often. The consequence of not changing theDMAP often is that the service delivery cost wouldbe high because of the triangular routing path CN-DMAP-MN for data communication between theCN and MN. On the other hand, a small servicearea size means that the DMAP will be changedoften so it will stay close to the MN. The conse-quence is that the communication cost for servicedata delivery would be low because of the shortCN-DMAP-MN route. However, a DMAP changeinvolves the cost of informing the HA and CNs ofthe RCoA address change. Therefore, there is atrade-off between these two cost factors and anoptimal service area exists.

The service and mobility characteristics of a MNare summarized by two parameters. The firstparameter is the residence time that the MN staysin a subnet. This parameter can be collected byeach MN based on statistical analysis [1]. We exp-ect that future MNs are reasonably powerful forcollecting data and doing statistical analysis. Theresidence time in general would be characterizedby a general distribution. Loosely, we use the MN’smobility rate (σ ) to represent this parameter. Thesecond parameter is the service traffic betweenthe MN and server applications. The MN can also

collect data statistically to parameterize this.Loosely, we use the data packet rate (λ) betweenthe MN and CNs to represent this parameter. Bothof these parameters are to be determined by theMN. For efficiency, the MN could build a tableto lookup its mobility and service rates as a func-tion of its location, time of the day, and day of theweek, based on statistical analysis. The ratio of λ/µ

is called the service to mobility ratio (SMR) of theMN.

When a MN moves across a service area bound-ary, the DMAP changes, thus incurring a “servicehandoff.” This service handoff cost includes the sig-naling cost to inform the HA and CNs of the newRCoA of the MN. In IPv6-based systems, the com-munication overhead between two communicatingprocesses is measured by the number of hops. Sincethe number of subnets separating two communi-cating processes would not properly measure thehop-count distance, we let F(K) denote a functionthat returns the number of hops as a function of thenumber of subnets K. This function F(K) can beperiodically and dynamically determined by a MNwhich collects statistical data as it roams across sub-nets. A MN would determine this function dynam-ically. The fluid flow model [13] assumes that theaverage number of hops between two communi-cating processes separated by K subnets is equalto

√K.

Table 1 lists a set of identified system parametersthat characterize the mobility and service charac-teristics of a MN in a MIPv6 system.

Our DMAP scheme is per-user based. Datapackets from a server application are sent to aMN’s DMAP first (which is just an AR selectedto be the current DMAP) and then forwarded tothe MN. The DMAP will receive packets addressedto the RCoA from the HA and CNs. Packets willbe tunneled from the DMAP to the MN.

3.1 Intra-regional move

When a MN performs a location handoff within aservice area, it acquires a CoA from the subnet andinforms the DMAP of the CoA address change.The DMAP is not changed. Also, the HA and CNare not informed of the CoA address change of theMN since they only know the MN by the RCoAaddress which is not changed in this case.


Table 1 Parameters for integrated mobility and service management

Symbol Meaning

λ Data packet rate between the MN and CNsσ Mobility rate at which the MN moves across subnet boundariesSMR Service to mobility ratio (λ/σ )N Number of server engaged by the MNF(K) A general function relating the number of subnets K to the number of hopsK Number of subnets in one service areaτ 1-hop communication delay per packet in wired networksα Average distance between HA and MAPβ Average distance between CN and MAPγ Cost ratio between wireless vs. wired network

3.2 Inter-regional move

When a MN moves across a service area thus incur-ring a service handoff, the MN acquires a CoAand a RCoA from the AR that now becomes theDMAP and an entry (RCoA, CoA) is recorded inthe lookup table of the AR. The MN then informsthe HA and CNs of its new RCoA to complete theservice handoff. The implementation of the pro-posed DMAP scheme based on the “DMAP tablelookup” design is totally transparent to the HAand CNs. The HA and CNs are informed of theRCoA as part of the service handoff process. Pack-ets from the HA or CNs will use the RCoA as thedestination address, which will be intercepted bythe AR that serves as the MN’s DMAP who willthen forward them to the MN. The packet rout-ing and forwarding will be done entirely in thenetwork layer. In effect the system behaves as ifa two-level HMIPv6 structure has been used todo both mobility and service management, exceptthat the service area is to be determined by theMN. Instead of having the MN discover the pres-ence of a MAP through announcement packetsand initiating the MAP migration process, the MNwill dynamically determine which AR will act as aDMAP to minimize the network cost.

4 Model

We devise a computational procedure to determinethe optimal service area size utilizing stochasticPetri net (SPN) techniques. The intent to find theoptimal service area based on the MN’s mobility

and service behaviors. The designer would utilizethe computational procedure developed here tobuild a table at static time listing the optimal ser-vice area as a function of these parameters eachcovering a reasonable value range. Such a table isthen loaded into the MN. The actual values of theseparameters are dynamically collected by the MN atruntime. Based on the values of these parameters atthe time a service area is crossed, the MN performsa simple table lookup to determine the optimal ser-vice area. The computational procedure requiresthat (1) every AR must be capable of acting as amobile anchor point (MAP); and (2) each MN mustbe powerful enough to collect data dynamically andperform simple statistical analysis to parameterizeits mobility and service characteristics.

The metric that we aim to minimize is the “com-munication cost” incurred per time unit due tomobility and service operations. The communica-tion cost includes the signaling overhead for mobil-ity management for informing the DMAP of theCoA changes, and informing the HA and CNs ofthe RCoA changes, as well as the communicationoverhead for service management for deliveringdata packets between the MN and CNs. Our SPNmodel is shown in Fig. 2. Table 2 gives the meaningof places and transitions defined in the model. Wechoose SPN because of its ability to deal with gen-eral time distributions for events, its concise repre-sentation of the underlying state machine to dealwith a large number of states, and its expressive-ness to reason about a MN’s behavior as it migratesfrom one state to another in response to eventsoccurring in the system. Once the parameters ofthe SPN model are given proper values, numerical


MovesXs

Move

MN2DMAP

NewDMAP

K

K

(Guard:mark(Xs)=K)

tmp

(Guard:Mark(Xs) < K-1

(Guard:Mark(Xs) = K-1)

Pj=1

Pi=1

A

B

Fig. 2 Stochastic Petri net model

Table 2 Meaning of places and transitions in the SPN model

Symbol Meaning

Move A timed transition for the MN to move across subnet areasMoves Mark(Moves) = 1 meaning that the MN just moves across a subnetMN2DMAP A timed transition for the MN to inform the DMAP of the CoA changeXs Mark(Xs) holds the number of subnets crossed in a service areaNewDMAP A timed transition to inform the HA and CNs of the RCoA changeK Number of subnets under a service areaGuard:Mark(Xs)< K − 1 A guard for transition A that is enabled if a move will not cross a service areaGuard:Mark(Xs)= K − 1 A guard for transition B that is enabled if a move will cross a service areaGuard:Mark(Xs)= K A guard for transition NewDMAP that is enabled if the MN just moves across a service areaA An immediate transition when Mark(Xs)< K − 1B An immediate transition when Mark(Xs)= K − 1tmp A temporary place that holds tokens from transition A

analysis methods for solving SPN models based onSOR or Gauss Seidel [11] are readily available tocompute the optimal service area size.

A token in the SPN model represents a subnetcrossing event by the MN. The function Mark(P)returns the number of tokens in place P. The num-ber of tokens accumulated in place Xs, that is,Mark(Xs), represents the number of subnetsalready crossed by the MN since it enters a newservice area. The SPN model describes thebehavior of a MN operating under the DMAPscheme:

1. When a MN moves across a subnet area, thusincurring a location handoff, a token is put inplace Moves. The mobility rate at which loca-tion handoffs occur is σ which is the transitionrate assigned to Move.

2. If the current move is an intra-regional movesuch that the guard for transition A willreturn true, then the MN will only inform the

DMAP of the CoA change. This is modeled byfiring immediate transition A, allowing the to-ken in place Moves to move to place tmp. Sub-sequently, once the MN obtains a CoA fromthe subnet it just enters, it will communicatewith the DMAP of the new CoA change. Thisis modeled by enabling and firing transitionMN2DMAP. After MN2DMAP is fired, a tokenin place tmp flows to place Xs, representing thata location handoff has been completed and theDMAP has been informed of the CoA changeof the MN.

3. If the current move results in the total numberof moves being K such that the guard for transi-tion B will return true, then the move will makethe MN cross a service area. This is modeled byenabling and thus firing immediate transitionB, allowing the token in place Moves to moveto place Xs in preparation for a service handoffevent. Note that in an SPN, firing an immediatetransition does not take any time.


4. If the number of moves, including the currentone, in place Xs has accumulated to K, a thresh-old determined by the DMAP representing thesize of a service area, then it means that theMN has just moved into a new service areaand a service handoff ensues. This is modeledby assigning an enabling function that will ena-ble transition NewDMAP when K tokens havebeen accumulated in place Xs. After transitionNewDMAP is fired, all K tokens are consumedand place Xs contains no token, representingthat the AR of the subnet that the MN justenters has been appointed as the DMAP bythe MN in the new service area.

Below we show an example of parameterizingtransition rates of MN2DMAP and NewDMAP basedon the set of base parameters defined in Table 1.The firing time of transition MN2DMAP stands forthe communication time of the MN informing theDMAP of the new CoA through the wireless net-work. This time depends on the number of hopsseparating the MN and its DMAP. Thus, the tran-sition rate of transition MN2DMAP is calculatedas:

1γ τ + F(Mark(Xs) + 1) × τ

,

where τ stands for the one-hop communicationdelay per packet in the wired network and γ isa proportionality constant representing the ratioof the communication delay in the wireless net-work to the communication delay in the wired net-work. F(Mark(Xs)+1) returns the number of hopsbetween the current subnet and the DMAP sepa-rated by Mark(Xs)+1 subnets. The argument of theF(x) function is added by 1 to satisfy the initial con-dition that Mark(Xs) = 0 in which the DMAP hasjust moved into a new service area, so at the firstsubnet crossing event, the distance between theDMAP and the subnet is one subnet apart. Notethat this transition rate is state-dependent becausethe number of tokens in place Xs changes dynam-ically over time. When transition NewDMAP fires,the AR of the subnet that the MN moves into willbe selected as the DMAP. The communication costincludes that for the MN to inform the HA and CNsof the new RCoA address change, i.e., (α + Nβ)τ ,where α is the average distance in hops betweenthe MN and the HA, β is the average distance

in hops between the MN and a CN, and N is thenumber of CNs that the MN concurrently engages.Thus, the transition rate of transition NewDMAP iscalculated as:

1(α + Nβ)τ

.

The stochastic model underlying the SPN modelis a continuous-time semi-Markov chain with thestate representation of (a, b) where a is the numberof tokens in place Moves, b is the number of tokensin place Xs. Let Pi be the steady state probabilitythat the system is found to contain i tokens in placeXs such that Mark(Xs)=i. Let Ci,service be the com-munication overhead for the network to service adata packet when the MN is in the ith subnet in theservice area.

Let Cservice be the average communication over-head to service a data packet weighted by therespective Pi probabilities. The communicationoverhead includes a communication delay betweenthe DMAP and a CN in the fixed network, a delayfrom DMAP to the AR of the MN’s current subnetin the fixed network, and a delay in the wireless linkfrom the AR to the MN. Thus, Cservice is calculatedas follows:

Cservice =K∑

i=0

(Pi × Ci,service)

= γ τ + βτ +K∑

i=0

(Pi × F(i)τ ). (1)

Let Ci,location be the network signaling overheadto service a location handoff operation given thatthe MN is in the ith subnet in the service area. Ifi < K, only a minimum signaling cost will incurredfor the MN to inform the DMAP of the CoA add-ress change. On the other hand, if i = K, then thelocation handoff also triggers a service handoff. Aservice handoff will incur a higher communicationsignaling cost to inform the HA and N CNs (orapplication servers) of the RCoA address change.Let Clocation be the average communication costto service a move operation by the MN weightedby the respective Pi probabilities. Then, Clocation iscalculated as follows:


Clocation =K∑

i=0

(Pi × Ci,location)

= PK(γ τ + ατ + Nβτ)

+K−1∑

i=0

{Pi(γ τ + F(K)τ )}. (2)

Summarizing above, the total communicationcost per time unit for the Mobile IP network operat-ing under our DMAP scheme to service operationsassociated with mobility and service managementof the MN is calculated as:

CDMAP = Cservice × λ + Clocation × σ , (3)

where λ is the data packet rate between the MNand CNs, and σ is the MN’s mobility rate. Thesteady-state probability Pi, 1 ≤ i ≤ K, needed inEqs. 1 and 2 can be solved easily utilizing numericalmethod solution techniques such as SOR or GaussSeidel [11].

5 Numerical results

Here we apply Eqs. 1–3 to calculate CDMAP as afunction of K and determine the optimal K rep-resenting the optimal “service area” size that willminimize the network cost, when given a set ofparameter values characterizing the MN’s mobil-ity and service behaviors. Below we present resultsto show that there exists an optimal service areafor systems operating under DMAP for networkcost minimization, and demonstrate the benefit ofDMAP over basic MIPv6 and HMIPv6.

For basic MIPv6, there is no DMAP. Thus, thecommunication cost CMIPv6

service for servicing a packetdelivery in basic MIPv6 includes a communication

delay from the CN to the AR of the current subnet,and a delay in the wireless link from the AR to theMN as follows:

CMIPv6service = γ τ + ατ . (4)

Under basic MIPv6, when a MN crosses a sub-net boundary, thus incurring a location handoff, theMN informs the HA and CNs of its CoA change.The cost CMIPv6

location for servicing a location handoffunder basic MIPv6 consists of a delay in the wire-less link from the MN to the AR of the subnet thatit just enters into, a delay from that AR to the CNs,and a delay from that AR to the HA as follows:

CMIPv6location = γ τ + ατ + Nβτ . (5)

The total cost per time unit for servicing datadelivery and mobility management operationsunder MIPv6 is given by:

CMIPv6 = CMIPv6service × λ + CMIPv6

location × σ . (6)

For HMIPv6, the placement of MAPs is pre-determined. We compare DMAP with an imple-mentation of HMIPv6 in which each MAP coversa fixed number of subnets, say, KH = 4. Thus, whena MN crosses a subnet within a MAP domain ofKH subnets, it only informs the MAP of its CoA.On the other hand, when the MN crosses a MAPdomain, it changes the MAP, obtains a new RCoAand informs the HA and CNs of the new RCoA.

Table 3 compares the communication costincurred per time unit by DMAP versus that by ba-sic MIPv6 and HMIPv6 head-to-head as afunction of SMR, from the perspective of Kopt thatminimizes the overall communication cost. Theexample is based on F(K) = √

K [13], α = β = 30,and γ = 10. The cost metric is normalized withrespect to τ = 1. First we observe that Kopt exi-sts under DMAP. Furthermore, as SMR increases,

Table 3 ComparingDMAP with basic MIPv6and HMIPv6head-to-head from theperspective of Kopt

SMR CMIPv6 CHMIPv6(KH = 4) CDMAP(Kopt) Kopt

0.1250 1.8750 0.7897 0.5522 340.2500 2.0000 0.9187 0.6892 320.5000 2.2500 1.1766 0.9619 281.0000 2.7500 1.6925 1.5034 222.0000 3.7500 2.7242 2.5758 164.0000 5.7500 4.7876 4.6960 118.0000 9.7500 8.9144 8.8859 7

16.0000 17.7500 17.1681 17.1681 432.0000 33.7500 33.6754 33.5475 264.0000 65.7500 66.6901 65.7500 1


Kopt decreases because when SMR is large, theservice rate is high compared with the mobilityrate, so the DMAP likes to stay as close to the MNas possible to minimize the communication over-head through the CN-DMAP-MN route. This tableclearly exhibits the behavior of DMAP. We see thatDMAP dominates basic MIPv6 when SMR is low.As SMR increases exceeding a threshold (e.g., 64),Kopt approaches 1 under which DMAP degener-ates to basic MIPv6. The reason is that when SMRis sufficiently high, the MN’s packet arrival rateis much higher than the mobility rate, so the datadelivery cost dominates the mobility managementcost. Therefore, the DMAP will stay close to theMN to lower the data delivery cost, thus makingKopt = 1 in our DMAP scheme in order to reducethe CN-DMAP-MN (or CN-MAP-MN) triangularrouting cost for packet delivery.

Comparing DMAP with HMIPv6 from the per-spective of Kopt, we observe from Table 3 thatDMAP degenerates to HMIPv6 (with a fixed KH)when SMR is moderate. When SMR is either highor low, DMAP performs substantially better thanHMIPv6. The trend would be true with otherchoices of KH for HMIPv6 as well (say KH = 8).When SMR is low, Kopt is high for DMAP com-pared with a fixed KH for HMIPv6, in which casethe cost saving is due to mobility cost reduction.On the other hand, when SMR is high, Kopt is lowfor DMAP compared with a fixed KH for HMIPv6,in which case the cost saving is due to service costreduction. Overall, there is a wide range of SMRvalues under which DMAP dominates HMIPv6because of the choice of Kopt for DMAP vs. a fixedKH for HMIPv6. Here we should note that a smalldifference of 0.1 is considered significant becausethe metric is expressed in terms of the networkcost per time unit (normalized to the per-hop costτ = 1) so the cumulative effect would be signifi-cant over the lifetime of a single MN. It is alsoworth noting that the cumulative effect of cost sav-ing for a large number of MN would be even moresignificant.

Figure 3 summarizes the cost difference betweenbasic MIPv6 and DMAP versus HMIPv6 andDMAP as a function of SMR. The Y coordinateis the cost difference incurred per time unit. Thereare two curves shown in Fig. 3. The first curveshows the cost difference between basic MIPv6

and DMAP (CMIPv6 − CDMAP), and the secondcurve shows the cost difference between HMIPv6and DMAP (CHMIPv6 − CDMAP), as a functionof SMR. We first observe that the cost differencebetween basic MIPv6 and DMAP (the first curve)decreases as SMR increases. The reason is thatDMAP degenerates to MIPv6 when SMR becomessufficiently large. We conclude that DMAP per-forms significantly better than basic MIPv6 espe-cially when SMR is low. Next we observe that thecost difference between HMIPv6 and DMAP (thesecond curve) initially decreases as SMRincreases until Kopt coincides with KH at whichpoint DMAP degenerates to HMIPv6, and thenthe cost difference increases sharply as SMR con-tinues to increase. We conclude that DMAP per-forms significantly better than HMIPv6 when SMRis either low or high.

Figure 4 tests the sensitivity of the results withrespect to α and β representing the hop distancesbetween the MN and the HA and CNs. We see thatas the hop distance between the HA (and CNs) inc-reases, the cost difference between HMIPv6 andDMAP (CHMIPv6 − CDMAP) becomes more pro-nounced, especially when SMR is small. The reasonis that when SMR is small, the mobility manage-ment cost dominates the data delivery cost, so Kopttends to be large to reduce the mobility manage-ment cost. In this case DMAP dictates a high Koptvalue to be used to reduce the mobility manage-ment cost as opposed to a fixed KH = 4 used byHMIPv6, thereby resulting in a more substantialcost difference between HMIPv6 and DMAP asSMR decreases. Recall that the cost is normal-ized with respect the per-hop communication costτ = 1. By observing the four curves in Fig. 4, nev-ertheless, we see that the trend remains the same.That is, the cost difference (CHMIPv6−CDMAP) dec-reases with the increase of SMR and then increasessharply after a threshold point at which DMAPdegenerates to HMIPv6. Figure 4 shows that undera wide range of α- and β- values, DMAP alwaysincurs less network overheads than HMIPv6, theeffect of which is pronounced when SMR is rela-tively low or high.

All the above results are based on the assump-tion that the average number of hops between theDMAP and MN separated by k subnets is given byF(k) = √

k, which are based on the fluid flow model


Fig. 3 Cost differencebetween basic MIPv6,HMIPv6, and DMAP

0

0.5

1

1.5

2

2.5

128 64 32 1684211/21/41/8

Cos

t Diff

eren

ce

SMR

CHMIPv6 - CDMAPCMIPv6 - CDMAP

Fig. 4 Effect of α and β

on CHMIPv6 − CDMAP

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

128 64 32 1684211/21/41/8

CH

MIP

v6 -

CD

MA

P

SMR

α=10, β=10α=20, β=20α=30, β=30α=40, β=40

[13]. We have tested the sensitivity to the form ofF(k). Figure 5 shows the effect of F(k) on the costdifference CHMIPv6 − CDMAP for F(k) = √

k andF(k) = k. The cost difference curves exhibited arevery similar in shape and are not sensitive to theform of F(k). We conclude that all the conclusionsdrawn earlier from the case F(k) = √

k are valid.

6 Applicability and conclusion

In this paper, we proposed a novel DMAP schemefor integrated mobility and service management.

The core of the idea lies in allowing intelligentMNs to determine the best service areas for ser-vice handoffs to minimize the network signalingcost due to mobility and service management. Wedevised a procedure to compute the optimal ser-vice area size that would minimize the overall com-munication cost, when given a set of parameterscharacterizing the mobile node’s mobility and ser-vice characteristics. We compared our scheme withbasic MIPv6 and HMIPv6. Our scheme outper-forms both basic MIPv6 and HMIPv6 in terms ofthe network communication overhead, the effectof which is especially pronounced when the service


Fig. 5 Effect of F(k) onCHMIPv6 − CDMAP

0

0.5

1

1.5

2

2.5

128 64 32 1684211/21/41/8

Cos

t Diff

eren

ce (

CH

MIP

v6 -

CD

MA

P)

SMR

f(k)=kf(k)=sqrt(k)

to mobility ration (SMR) is low for basic MIPv6,and either low or high for HMIPv6.

To apply the analysis results in the paper, onecan execute the computational procedure at sta-tic time to determine optimal Kopt over a possi-ble range of parameter values. Then a MN canperform a simple look-up operation to determineKopt based on data collected at runtime. This all-ows each MN to dynamically determine the bestservice area size to minimize the overall communi-cation cost. The performance gain is in the amountof communication cost saved per time unit per user,so the saving due to a proper selection of the bestservice area will have significant impacts since thecumulative effect for all mobile users over a longtime period would be significant.

In the future, we plan to consider the implemen-tation issue by building a testbed system to validatethe analytical results as well as testing the sensitiv-ity of the results with respect to other time distribu-tions other than the exponential distribution usedin the analysis. This testbed consists of DMAP-compliant MNs, ARs, and server applications in aMobile IPv6 environment, leveraging existing codefor HMIPv6 running on Linux platforms [8]. Wewill test scenarios with variations in the mobilityand service characteristics of MNs in this testbedto compare DMAP with MIPv6 and HMIPv6 forexperimental validation. We also plan to extendthe research to the case in which a user proxy can

be used to execute the function of DMAP for data-base applications that use per-user, stateful proxiesto effectively cache data items to support discon-nected operations to further reduce the networkcommunication cost.

References

1. Chen, I. R., & Verma, N. (2003). Simulation study ofa class of autonomous host-centric mobility predic-tion algorithms for cellular and ad hoc networks. InProceedings of the 36th annual simulation symposium,pp. 65–72. Orlando, USA.

2. Dimopoulou, L., Leoleis, G., & Venieris, I.S. (2004).Introducing a hybrid fast and hierarchical MIPv6scheme in a UMTS-IP converged architecture. In pro-ceedings of the 29th IEEE international conference onlocal computer networks, pp. 42–49. Tampa, USA.

3. Gu, B., & Chen, I.R. (2005). Performance analysis oflocation-aware mobile service proxies for reducing net-work cost in personal communication systems. ACMMobile Networks and Applications, 10(4), 453–463.

4. Gustafsson, E., Jonsson, A., & Perkins, C. (2006).Mobile IPv4 regional registration. http://www.ietf.org/internet-drafts/draft-ietf-mip4-reg-tunnel-02.txt, IETF,Work in Progress.

5. Johnson, D., Perkins, C., & Arkko, J. (2004). Mobil-ity support in IPv6. http://www.ietf.org/rfc/rfc3775.txt,IETF, Work in Progress.

6. Misra, A., Das, S., Dutta, A., McAuley, A., & Das,S.K. (2001). IDMP-based fast handoffs and pagingin IP-based cellular networks. In Proceedings of theInternational conference on 3G wireless and beyond,pp. 427–432. San Francisco, USA.


7. Misra, A., Das, S., Dutta, A., McAuley, A., & Das, S.K.(2000). IDMP: An Intra-Domain Mobility ManagementProtocol using Mobility Agents. draft-mobileipmisra-idmp-00.txt, IETF, Work in Progress.

8. Nelson, R., Daley, G., & Moore, N. (2002). Imple-mentation of hierarchical mobile IPv6 for linux. InProceedings of the 6th international symposium on com-munications interworking (IFIP Interworking 2002).Perth, Australia.

9. Pack, S., Choi, Y., & Nam, M. (2006). Design and anal-ysis of optimal multi-level hierarchical Mobile IPv6networks. Wireless Personal Communications, 36(2)95–112.

10. Soliman, H., Castelluccia, C., El-Malki, K., & Bellier,L. (2005). Hierarchical mobile IPv6 mobility manage-ment. http://www.ietf.org/rfc/rfc4140.txt, IETF, Work inProgress.

11. Trivedi, K.S., Ciardo, G., & Muppala, J. (1999). SPNPVersion 6 User Manual. Dept. of Electrical Engineering,Duke University, Durham, NC.

12. Xie, J. & Akyildiz, I.F. (2002). A novel distributeddynamic location management scheme for minimizingsignaling costs in Mobile IP. IEEE Transactions on Mo-bile Computing, 1(3), 163–175.

13. Zhang, X., Castellanos, J., & Campbell, A. (2002).P–MIP: Paging extensions for Mobile IP. ACM MobileNetworks and Applications, 7(2), 127–141.

Ing-Ray Chen receivedthe BS degree fromthe National TaiwanUniversity, Taipei, Tai-wan, and the MS andPhD degrees in com-puter science from theUniversity of Hou-ston. He is a professorin the Department ofComputer Science atVirginia Tech. His res-earch interests includemobile computing, per-vasive computing, mul-

timedia, distributed systems, real-time intelligent systems,and reliability and performance analysis. Dr. Chen hasserved on the program committee of numerous conferences,including as program chair for 29th IEEE Annual Interna-tional Computer Software and Application Conference in2005, 14th IEEE International Conference on Tools withArtificial Intelligence in 2002, and 3rd IEEE Symposiumon Application-Specific Systems and Software EngineeringTechnology in 2000. Dr. Chen currently serves as an edi-tor for Wireless Personal Communications, The ComputerJournal, and International Journal on Artificial IntelligenceTools. He is a member of the IEEE/CS and ACM.

Weiping He receivedhis BE degree fromHuangzhong Univer-sity of Science andTechnology, China, in1996, and his Masterdegree in ComputerScience from VirginiaTech in 2002. He isa doctoral candidatein the Department ofComputer Science atVirginia Tech. His res-earch interests includefuture mobile commu-

nication system architectures, design and evaluation ofmobility and service management schemes in mobilecomputing environments.

Baoshan Gu receivedthe BS degree fromUniversity of Scienceand Technology ofChina, Hefei, China,in 1992 and the MSdegree in computerscience from Insti-tute of ComputingTechnology, ChineseAcademia of Science,Beijing, China, in 1995.He received his PhDdegree in ComputerScience from Virginia

Tech in 2005. From 1995 to 2000, he was a research anddevelopment engineer in Institute of Computing Technol-ogy, Chinese Academia of Science. Since 2006, he has beena research scientist at NIH. His research interests includenext-generation wireless system architectures, design andevaluation of location and service management schemes inmobile computing environments, and mobile multimediasystems.

DMAP: integrated mobility and service management in mobile IPv6

Documents

Transcript of DMAP: integrated mobility and service management in mobile IPv6