A Local Mobility Agent Selection Algorithm for Mobile Networks Yi Xu

Henry C. J. Lee

Vrizlynn L. L. Thing

Institute for Infocomm Research, 21 Heng Mui Keng Terrace, Singapore 119613 Email: {yxu, hlee, vriz}@i2r.a-star.edu.sg Abstract— The Mobile IP protocol has been designed to address the problem of roaming between IP networks. However, as a mobile node moves between networks, the signaling overhead causes significant disruption to real time data traffic. The Localized Mobility Management (LMM) has been proposed to enhance the handoff performance between networks within the same domain. By introducing the concept of Local Mobility Agent (LMA), a mobile node visiting a foreign domain is exempted from sending frequent address update to its home agent and correspondent nodes, when its movement is limited to the visited domain. As multiple LMAs are configured in a domain for redundancy, scalability and load sharing concerns, LMA selection becomes an issue. This paper proposes a new LMA selection algorithm, Mobile Controlled Movement Tracking (MCMT). The objective of this new algorithm is to discover the optimal LMA in terms of selection stability and load balancing, by taking into consideration the mobility characteristics of the mobile node. Our analysis shows that this proposal can provide good support for low latency handoff and load sharing among LMAs. Index Terms—Mobile IP, Localized Mobility Management, Local Mobility Agent, Internet, Wireless Networks.

I. I NTRODUCTION Mobility support in contemporary internetworks is a fundamental issue when these networks promise location independent service. No matter where a host resides physically, the network should be able to identify it correctly with transparent support for communications above network layer. Mobile IP has been proposed to address the mobility issue in IP layer [1, 2]. When away from the home network, a mobile node registers its new IP address (care of address) with the home agent. The correspondent nodes, which are unaware of the new location of the mobile node, send packets to its home network as usual. The home agent intercepts these packets and redirects them to the current care of address of the mobile node through tunneling. Meanwhile, the mobile node may update the correspondent nodes with its new address so that packets from the correspondent nodes can be routed directly to the mobile node after address update. While mobile IP allows ubiquitous accessibility to a mobile node, performance in upper layer application support is not so satisfactory, especially for QoS sensitive communications. In basic mobile IP proposal, each time the mobile node moves into a new access network, a sequence of signaling takes place before upper layer communication session is restored into normal transmission. During this interim, the mobile node completes link layer handoff, detects network alteration, configures its new IP address, and sends binding updates to its home agent and correspondent nodes. Handoff latency in IP layer may cause noticeable disruption to ongoing transmission session that all the packets sent to the mobile node are lost dur-

ing this interval. Congestion control mechanism of TCP degrades throughput further by mistakenly reducing the transmission window size, having observed packet loss. The basic mobile IP proposal also incurs high overhead on the network as every binding update message must traverse the network all the way back to the home agent and the correspondent nodes even if the mobile node is involved only in local movement within a domain. These performance problems for mobile IP need to be addressed especially in the wireless environment. Deployment of small and dense wireless cell offers high aggregate bandwidth, but forces frequent handoff for a moving node. To accommodate the mobile node with better support for seamless handoff and alleviate network load as much as possible, extensions to the basic mobile IP are proposed. In IPv4 networks, the foreign agents are organized into a treelike hierarchy to improve mobile IP handoff latency. In [3], all the foreign agents on the path from the mobile node to its home network keeps an address binding for the mobile node. Each binding points to the next hop leading to the present location of the mobile node. When the mobile node changes location, the address update message stops at the foreign agent where a binding already exists, as the remaining hops are intact from the address change. In [4], different mobility management strategies are employed respectively for local, intra administrative domain and global mobility. While local mobility is taken care of by link layer mechanism and global mobility is handled by the basic mobile IP, hierarchy of foreign agents is assumed to manage intra-domain movement of the mobile node. In IPv6 networks, the concept of Local Mobility Agent (LMA) [5] is introduced that serves as a local home agent in the visited domain. In [6] the Mobility Anchor Point (MAP) is used as a terminology equivalent to LMA. By designating LMA within a domain visited by the mobile node, intra-domain movement of the mobile node is hidden behind the LMA so that its handoff is transparent to the home agent and the correspondent nodes. Each time the mobile node changes its point of attachment, it sends a binding update message to the LMA, whereas its globally perceived IP address remains intact and valid for routing outside the domain. Therefore, handoff latency is reduced as binding update terminates at the LMA in local domain rather than at the home agent and correspondent nodes which may reside at the other end of Internet. Overhead introduced by location management is also lower than that of the basic mobile IP. Instead of separate binding update with the home agent and each correspondent node, one update with the LMA is enough to track the movement of the mobile node. Other approaches to reduce IP handoff latency and packet loss include fast handoff [7], utilization of IP multicasting [8,9] and link layer handoff notification to IP layer [10].

0-7803-7802-4/03/$17.00 © 2003 IEEE

1074

In this paper we study the localized mobility management approach in IPv6 networks. To improve robustness and enable traffic sharing, multiple LMAs are typically deployed in a domain. Normally a mobile node does not select more than one LMA, because the use of additional LMA introduces extra tunneling overhead when packets are routed to the mobile node. This paper attempts to address the issue of selecting one LMA in a domain with multiple LMAs. The rest of this paper is organized as follows. Section II describes the related work and states the problem to be studied. Section III proposes a new LMA selection algorithm, the Mobile Controlled Movement Tracking. Section IV investigates this new proposal with analytical and simulation results. Section V concludes this paper. II. R ELATED W ORK AND P ROBLEM S TATEMENT Two LMA selection algorithms are seen in literature. One selects the furthest LMA on the transmission path from the mobile node to its home network [6, 11] and the other makes selection by the preference set by each LMA [6]. An IPv6 neighbor discovery extension, the MAP option, is defined in [6] to disseminate LMA information throughout a domain. Two informational fields are included in this option for LMA selection. The Distance field records the distance from the LMA to the mobile node. The Preference field indicates the willingness of the LMA to offer local registration service. Every LMA inside the domain keeps sending out its option to the mobile node periodically. For the furthest LMA selection, the mobile node chooses to register with the LMA with the greatest Distance field. Because this one appears to be always reachable in the long run, this selection minimizes frequent registration at the home agent and the correspondent nodes when the mobile node changes its care of address. For the preference selection, the mobile node selects the LMA with the smallest value in the Preference field. However, the furthest LMA selection strategy has some drawbacks. Firstly, the furthest LMA from the mobile node will most likely be the gateway that connects the domain to the rest of the networks. If every mobile node selects the furthest LMA, this furthest one is likely to be overloaded as it is the sole entity responsible for all the mobile nodes in the domain. Secondly, if the mobility scope of the mobile node is only a limited area in the domain, it is not necessary to register at the furthest LMA. In this case, a nearer LMA will reduce registration delay. For the preference based LMA selection, the unresolved problem is what factor decides value assignment for preference level. We propose that LMA selection follow the principle that the most suitable LMA is the one which provides not-so-transient regional registration service and meanwhile resides near to the mobile node. For a mobile node that changes access network frequently, a relatively further LMA will take care of its mobility, while for a mobile node that is involved in limited mobility, a relatively nearer LMA is selected. In this way, different mobile nodes may end up with different LMA selection. Consequently, registration load is distributed to all the LMAs in the

Lifetime entry 1 (Gateway LMA) entry 2

entry n

Distance IP Address

Distance IP Address

Distance IP Address

Fig. 1. LMA advertisement, a new IPv6 neighbor discovery option

domain, avoiding overcrowding at the gateway LMA. Selection of a nearer LMA also brings another performance advantage that update of new care of address at the selected LMA can be completed faster. III. M OBILE C ONTROLLED M OVEMENT T RACKING LMA S ELECTION A new LMA selection algorithm, Mobile Controlled Movement Tracking (MCMT), is presented in this section. This name implies two facts. Firstly, while network provides the information on LMA distribution to the mobile node, it is the mobile node itself that decides with which LMA to perform local registration. Secondly, the decision is based on the movement of the mobile node so that the selected LMA is near to the mobile node and provides stable service despite the movement of the mobile node. The MCMT operation requires the mobile node to know all the available LMAs on its transmission path towards the home agent, so as to select the most appropriate one. To achieve this, we propose to set up a LMA tree topology explicitly and distribute relevant branch information to the mobile node. Therefore, in addition to LMA selection, the MCMT strategy also includes LMA tree topology construction mechanism to configure all the LMAs of a domain into a tree structure. A. LMA Tree Construction We define a new IPv6 neighbor discovery option named LMA Advertisement option to construct the LMA tree. This option circulates with router advertisement messages to propagate LMA information throughout the domain. As illustrated in Figure 1, it contains three types of informational fields, Lifetime, Distance and IP Address. The Lifetime field indicates validity duration of the originating LMA. The Lifetime field is followed by a number of LMA entries, each comprising the (Distance, IP Address) pair. The IP Address field gives the IP address of the LMA. The Distance field records the distance from the LMA of this entry to the LMA that receives this advertisement. Meanwhile, each LMA maintains a LMA list locally that is a copy of the LMA advertisement. This LMA list lets each LMA know its own hierarchical position. The gateway LMA initiates the tree construction process by sending out the first LMA advertisement. This advertisement

1075

contains only one entry. Upon reception of this LMA advertisement at a downstream router, if this router does not act as LMA, it simply increments the Distance field by 1 and forwards this advertisement onto other network interfaces. If this router is a LMA, it examines whether a local LMA list already exists. If not, the list is established using the information carried in this advertisement. If a local list exists already, comparison is made between the distance recorded in the received advertisement and that in the local list to find out which one represents a shorter path from the gateway LMA. When the advertisement carries a shorter path, the local list is updated with the content of the advertisement. Otherwise, the local list remains intact. After a LMA has updated its local list, this LMA will generate a new advertisement to inform its neighbors of this change. In the new advertisement, this LMA adds in itself as the last entry and increments the Distance fields of all the other LMAs by 1. When the LMA advertisement propagates away from the gateway LMA, more LMAs are discovered and the length of the advertisement increases gradually until the access router is reached. Assuming that none of the LMA advertisements is lost, every LMA will eventually establish a local list that represents the shortest path from the gateway LMA. Each access router also keeps a local LMA list, which includes all the LMAs on the shortest path from the gateway LMA to the access router itself. The Lifetime field helps to detect network change, such as LMA failure and link disruption. Every LMA sends out advertisement periodically so that all its neighbors are continuously informed of its reachability. If advertisement from a LMA is absent for a time duration indicated by the Lifetime field in the last received advertisement, this LMA is deemed to be unreachable. All those affected LMAs will start to listen to other advertisements to find out another shortest LMA list.

for the mobile node to locate a new LMA different from the gateway LMA. Then the mobile node registers its local care of address at the newly selected LMA and registers the IP address of this newly selected LMA at the home agent and correspondent nodes. After that the mobile node starts a new round of search again. When the mobile node moves beyond the coverage of the selected LMA, it will immediately detect this by noticing that the serving LMA is absent in the latest received LMA list. A new LMA must be selected from the latest LMA list to take over the previous serving one. The mobile node identifies the new LMA to be the one that is almost the same distant away as the previous serving LMA. If mobility does not change after location transfer, this peer LMA is the most appropriate choice, as the mobility has already been correctly reflected by the previous selection. However, a new search still starts immediately after this new LMA is selected. If mobility actually changes after location transfer, at the end of this search interval either a further LMA or a nearer LMA will be located, depending on the new mobility pattern. When the mobile node changes LMA selection, the registration of new LMA IP address at the home agent and correspondent nodes may result in long delay. Before registration message is received, all packets sent out from the home agent and correspondent nodes are directed to the previous LMA and lost there. To reduce packet loss, each time the mobile node selects a new LMA, it also registers the new LMA IP address at previous LMA to redirect the in-flight packets to the new LMA. This registration is not permanent, as the in-flight packet stream lasts for only a short duration approximately equal to the round trip time from the mobile node to the home agent and correspondent node. Practically, this registration can expire automatically in a few seconds, which should already be long enough so that most in-flight packets will not be lost.

B. LMA Selection Access router periodically sends out LMA list together with router advertisement. From this list, the mobile node can find all the LMAs on the path from the gateway to the access router, as well as their respective distance. The mobile node can also request the LMA list from the access router actively instead of waiting for unsolicited announcement. When the mobile node enters a new domain, it selects the gateway LMA by default. The mobile node registers with the gateway LMA using local care of address and registers with the home agent and correspondent nodes using the IP address of the gateway LMA. Then the mobile node begins to search for a nearer LMA to replace the gateway LMA, if a nearer LMA is more suitable for its mobility pattern. For this purpose, the mobile node records all the LMA advertisements received during a predefined period called the search interval. At the end of search interval, the mobile node finds out the LMAs that have appeared in every LMA advertisement and the one nearest to the mobile node is selected to replace the gateway LMA. In case that the mobile node is involved in limited mobility within a fraction of the domain or it moves very slowly, it is possible

IV. P ERFORMANCE A NALYSIS Next, we study the performance of MCMT as compared to the furthest LMA selection. Discussion mainly focuses on two performance aspects, registration delay and load balancing. A. Registration Delay When the mobile node changes access network and registers the new care of address at LMA, shorter signaling delay can be expected when a nearer LMA is selected. If the mobile node is communicating with any correspondent node during this interval, the registration delay can be translated into packet loss. Intuitively, less packet loss will be resulted from selection of a nearer LMA than the furthest one. We study this performance aspect by simulation. The Network Simulator (NS-2) [12] is used as simulation tool. The simulation topology is depicted in Figure 2. In real-world scenario, there might be intermediate networks between neighboring LMAs. In our simulation topology, all these intermediate

1076

Packet Loss Ratio in Every 1−Minute Time Slice

CN 100Mbps 200ms

LMA 0

(Gateway LMA) 100Mbps 8ms

LMA 1 100Mbps 6ms

100Mbps 6ms

LMA 3

LMA 2 100Mbps 4ms

100Mbps 4ms

LMA 4 100Mbps 2ms

LMA 5

100Mbps 100Mbps 2ms 2ms

LMA 9

LMA 8 5Mbps 1ms

100Mbps 4ms

LMA10

100Mbps 4ms

LMA 6 100Mbps 100Mbps 2ms 2ms

LMA11

LMA12

LMA 7 100Mbps 100Mbps 2ms 2ms

LMA13

LMA14

100Mbps 2ms

LMA15

MN

Mobility Scope 1 Mobility Scope 2 Mobility Scope 3

1

Mobility Scope 1 Mobility Scope 2 Mobility Scope 3

0.9

0.8

0.7 0.6 0.5 0.4

0.3 0.2 0.1

0 0

Constant Movement

2

6 4 Simulation Time (min)

8

10

Fig. 3. Packet loss ratio in different mobility scopes, 4 min search interval

Fig. 2. Simulation topology

1

Then, we let the mobile node move in constant speed without direction change to investigate the effect of movement speed. The topology map is wrapped around so that when the mobile

0.9 0.8

Packet Loss Ratio

networks are reduced to a single link. Therefore, the delay assigned to each link includes propagation delay, transmission delay and processing delay on all the networks in between. Considering the routers near gateway are likely to be loaded more heavily than those near the access networks, we assign the delay values for each link in a decreasing order from the gateway LMA to the access router to reflect such difference in processing delay. For all wired links, bandwidth of 100Mbps is assumed. Because the last hop to the mobile node is wireless link that might be a bottleneck on the path, bandwidth of 5Mbps is assumed for the link between the mobile node and the nearest LMA. The correspondent node keeps sending in packets at constant rate of 1Mbps with packet size of 128 bytes each. When handoff takes place, we assume that the mobile node requests the LMA list from the new access router immediately. A delay of 10ms is used to model the LMA list acquisition process. Firstly, we let the mobile node move in different mobility scopes. For all three mobility scopes, the mobile node changes access network every 30 seconds. In scope 1, handoff happens between LMA8 and LMA9. In scope 2, handoff occurs between LMA8, LMA9, LMA10 and LMA11 in turn, and then in reverse order. In scope 3, handoff occurs from LMA8 to LMA15, then in reverse order. A search interval of 4 minutes is used. For each mobility scenario, packet loss in every 1-minute time slice is counted for both the MCMT algorithm and the furthest LMA selection. Figure 3 gives the packet loss ratio of the two strategies. As in the beginning 4 minutes MCMT selects the furthest LMA as default, there is no performance gain over the furthest LMA selection strategy. At the end of 4 minutes, depending on the respective mobility scope, MCMT locates a nearer LMA to replace the furthest one. After that, packet loss during handoff is reduced under MCMT algorithm. The most significant packet loss reduction occurs in mobility scope 1, because this one selects the nearest LMA among the three cases.

0.7

0.6

0.5

0.4

0.3 0.2

0.1

0 0

1 7 2 8 3 6 4 5 Dwelling Time Under Coverage of the Nearest LMA (min)

Fig. 4. Packet loss ratio in different speeds, 4 min search interval

node moves out of the coverage of LMA15, it enters the map again at LMA8. Dwelling time under coverage of the nearest LMA is used as equivalent measurement for speed. Figure 4 shows the packet loss ratio as a function of different dwelling time values. When the mobile node moves very fast, say, dwelling time is less than 1 minute, at the end of search interval, the mobile node goes across more than half of the coverage of LMA1, so LMA1 is decided to replace the gateway LMA. As dwelling time increases, the mobile node will select even nearer LMA gradually, so the packet loss ratio steps down. When dwelling time is longer than 4 minutes, search interval will always select the nearest LMA and the packet loss ratio reaches the lowest value. The packet loss ratio is determined by a few factors. Besides the local care of address registration delay and LMA list acquisition delay, each time the mobile node changes LMA, the signaling delay of new LMA IP address registration at previous LMA causes some additional packet loss. Comparison between Figure 3 and Figure 4 shows this phenomenon. In Figure 3, when the mobile node moves in the mobility scope 2, LMA2 is selected. In Figure 4, when the mobile node moves with dwelling time between 1 and 2 minutes, it selects LMA2 too. But in this latter case, the mobile node will switch between LMA2 and LMA3, therefore resulting higher packet loss ratio

1077

TABLE I

First, every n LMAs of the same distance from the mobile node are under coverage of the same parent LMA. In Figure 2, n is equal to 2. Second, all mobile nodes move constantly with the dwelling time under coverage of the nearest LMA uniformly distributed over [0, Tmax ]. According to the MCMT algorithm, when the dwelling time falls into the interval [s, Tmax ], the nearest LMA will be selected, where s denotes the length of search interval. When dwelling time falls into the interval [ ns , s], the second nearest LMA will be selected, and so on. The selection is summarized into the following equation.
Nearest LMA s < tdwl < Tmax (1) s s kth nearest LMA nk−1 < tdwl < nk−2 ,k ≥ 2

D ESCRIPTION OF A TYPICAL MOVEMENT SCENARIO

Packet Loss Ratio in Every 10−Minute Time Slice

Duration 30 min 30 min 30 min 30 min 30 min 30 min 30 min

Movement Description mobility scope 1, 50 sec dwelling time constant movement, 40 sec dwelling time mobility scope 1, 50 sec dwelling time constant movement, 20 sec dwelling time mobility scope 1, 50 sec dwelling time constant movement, 10 sec dwelling time mobility scope 1, 50 sec dwelling time

1

0.9

Known from Equation 1, when k ≥ 2, the number of mobile s s − nk−1 ), and nodes that select the kth nearest LMA is ρ( nk−2 s that select the (k + 1)th nearest LMA is ρ( nk−1 − nsk ), where ρ is the probability density. The ratio of the two groups are given in Equation 2. s s − nk−1 ) ρ( nk−2 =n (2) s s ρ( nk−1 − nk )

0.8

0.7 0.6

0.5

0.4

0.3

0.2

0.1 0 0

30

60

90 120 150 Simulation Time (min)

180

210

Fig. 5. Packet loss ratio in different time slice, 1 min search interval

than in Figure 3. Seen in Figure 3, the packet loss ratio is 0.38 after LMA2 has been selected, whereas the ratio is 0.46 in Figure 4. Nevertheless, as the signaling delay for registration at previous LMA is much shorter than the round trip time from the mobile node to the correspondent node, most in-flight packets can be redirected successfully and the packet loss ratio is still always smaller than 1. Therefore, even if the mobile node moves irregularly and goes beyond coverage of the currently selected LMA, the MCMT algorithm still does not perform worse than the furthest LMA selection. Finally, we present the simulation result of a more typical movement scenario. Table I describes this movement scenario. The result is depicted in Figure 5. As the mobile node changes its mobility pattern over time, LMA selection changes accordingly, so the packet loss ratio varies over time. Each ratio conforms well to our discussion on the previous two figures. B. Load Balancing Because each mobile node bases its LMA selection on its own mobility feature, intuitively, MCMT distributes the registration load to all the LMAs inside the domain. Therefore, better load balancing is achieved compared to the furthest LMA selection. However, the length of search interval plays an important role in LMA selection. To balance the load as evenly as possible among all the LMAs regardless their distance, a special case is studied, in which we make the following two assumptions.

As the number of LMAs in the kth nearest level is n times the number in the (k + 1)th nearest level, Equation 2 indicates that the average load on individual LMA is the same on both levels. Therefore, to balance the load on all levels, we only need to take care of the nearest LMAs. Let the average load on individual LMA in the nearest level be the same as that in the second nearest level, we have Equation 3. ρ(Tmax − s) s = ρ(s − ) n n

(3)

Then the search interval is determined by Equation 4. s=

Tmax n

(4)

In addition, if mobile nodes enter the domain at a constant arrival rate, then every LMA on the same distance level will also be evenly loaded, so that all the LMAs inside the domain will have the same load. In more realistic mobility scenarios, most mobile nodes do not move in high speed all the time. Instead, they tend to stay somewhere and change their location occasionally. In this case, most mobile nodes will select the LMAs in the lower levels of the LMA hierarchy. As the number of these LMAs far exceeds that near the gateway LMA, the major portion of registration load is therefore distributed, alleviating the load on the gateway LMA. C. Overhead Discussions The LMA tree construction and LMA advertisement messages introduce signaling overhead. However, this overhead does not exceed that introduced by the furthest LMA selection proposed in [6]. In [6], the LMA information is propagated in a similar way, and furthermore, a mobile node probably receives

1078

the advertisements from all the LMAs in the domain. But in the MCMT strategy, only the advertisements on its registration path are received. Therefore, the number of LMA advertisements transmitted on the wireless link is reduced. If the mobile node can request LMA list actively, the signaling overhead on the wireless link can be reduced even more. In this case, the LMA list is delivered to the mobile node only at request when handoff takes place, eliminating unnecessary announcement when the mobile node does not change access network. D. Comparison with Other Micro-Mobility Protocols There are many micro-mobility protocols seen in the literature. Besides deployment of local mobility agent in a domain, other techniques, such as fast handoff [7] and IP multicasting [8, 9], have also been discussed widely. The MCMT algorithm is orthogonal to them and, therefore, can be integrated with these techniques to reduce packet loss further. Among the approaches that assume employment of mobility agent, sequential tunneling along multiple mobility agents [3] and host specific routing [13, 14] are the two major methods proposed to track the local mobility of the mobile node. However, both of them require routing information to be recorded in multiple nodes, the mobility agents in the former one and the routers in the latter one. On the contrary, in the MCMT only the selected LMA is involved, which introduces less routing expense. Another problem with sequential tunneling and host specific routing is that the gateway mobility agent or router must keep a routing entry for every visiting mobile node, while the MCMT can potentially decentralize the load from the gateway. The storage space and messaging required by the MCMT for LMA tree construction are extra expenses that are not present in the other protocols. However, this overhead is independent of the number of mobile nodes in the domain. In contrast, in sequential tunneling and host specific routing, the number of routing entries in mobility agents and routers increases quickly when the mobile node population expands. Intuitively, there is less scalability concern in the MCMT than in the other two.

R EFERENCES [1] C. Perkins, ”IP Mobility Support for IPv4,” IETF RFC3220, Nokia Research Center, January 2002. [2] D. B. Johnson, C. Perkins, ”Mobility Support in IPv6,” IETF Internet Draft, work in progress, Nokia Research Center, March 2002. [3] D. Forsberg, J. T. Malinen, J. K. Malinen, T. Weckstrom, M. Tiusanen, ”Distributing Mobility Agents Hierarchically under Frequent Location Updates,” in Proceedings of Mobile Multimedia Communications 1999. [4] R. Caceres, V. N. Padmanabhan, ”Fast and Scalable Handoffs for Wireless Internetworks,” ACM MOBICOM 1996. [5] C. Williams, ”Localized Mobility Management Requirements for IPv6,” IETF Internet Draft, work in progress, DoCoMo USA Labs, November 2001. [6] H. Soliman, C. Castelluccia, K. El-Malki, L. Bellier, ”Hierarchical MIPv6 Mobility Management,” IETF Internet Draft, work in progress, Ericsson, July 2001. [7] G. Dommety, ”Fast Handovers for Mobile IPv6,” IETF Internet Draft, work in progress, March 2002. [8] C. Castelluccia, ”A Hierarchical Mobility Management Scheme for IPv6,” IEEE International Symposium on Computers and Communications 1998. [9] H. Balakrishnan, S. Seshan, R. H. Katz, ”Improving Reliable Transport and Handoff Performance in Cellular Wireless Networks,” ACM Wireless Networks, pp. 469-481, issue 1, 1995. [10] C. S. Wu, C. W. Cheng, N. F. Huang, G. K. Ma, ”Intelligent Handoff for Mobile Wireless Internet,” ACM Mobile Networks and Applications, pp. 67-79, issue 6, 2001. [11] V. L. L. Thing, H. C. J. Lee, Y. Xu, ”Design and Analysis of IPv6 Local Mobility Agents Discovery, Selection and Failure Detection,” The Fourth IEEE Conference on Mobile and Wireless Communications Networks, September 2002. [12] The Network Simulator, http://www.isi.edu/nsnam/ns. [13] A. T. Campbell, J. Gomez, S. Kim, A. G. Valko, C. Y. Wan, Z. R. Turanyi, ”Design, Implementation, and Evaluation of Cellular IP,” IEEE Personal Communications, pp. 42-49, August 2000. [14] R. Ramjee, T. La Porta, S. Thuel, K. Varadhan, ”HAWAII: A Domain-based Approach for Supporting Mobility in Wide-area Wireless Networks,” IEEE International Conference on Network Protocols, pp. 283-292, 1999.

V. C ONCLUSIONS A new LMA selection algorithm, Mobile Controlled Movement Tracking, has been proposed in this paper, which takes mobility pattern into consideration for LMA selection. By constant monitoring of node mobility, this algorithm attempts to locate a LMA that is stable and as near as possible to the mobile node. As a result, the selected LMA provides support for low latency handoff and load balancing. When mobility characteristics change over time, this algorithm adaptively discovers new LMA that is more suitable for the new mobility pattern. Simulation and analytical results show this new algorithm outperforms the furthest LMA selection in terms of reduced registration delay and balanced registration load. Compared to other micro-mobility management approaches, this proposal is also more scalable to support large number of mobile nodes.

1079