A Network-based Seamless Handover Scheme for Multi-homed Devices

Transcription

1 A Network-based Sealess Handover Schee for Multi-hoed Devices Md. Shohrab Hossain and Mohaed Atiquzzaan School of Coputer Science, University of Oklahoa, Noran, OK 7319 Eail: {shohrab, Abstract Terinal-based obility protocols require obile devices to participate in obility signaling that consues lots of processing power and eory. Network-based obility protocol solves this proble by excluding low-end obile devices fro signaling requireent. We earlier proposed SEMO6, a terinal-based obility protocol which exploits ultiple network interfaces to achieve sealess handover. In this paper, we have proposed a network-based obility solution for SEMO6. We have perfored a thorough cost analysis of its ajor obility entities that participate in obility anageent and also obtained their efficiency based on signaling overhead. Results show interesting relationships aong various network paraeters. Our signaling analysis can be used by network engineers to estiate the resource requireents of its entities in actual deployent. Index Ters Mobility anageent, Proxy Mobile IPv6, ultihoing, sealess handover. I. INTRODUCTION Proliferation in obile coputing has drawn significant attention of the research counity and Mobile IPv6 MIPv6) [1] was standardized to facilitate Internet connectivity to obile devices. However, as a host-based obility solution, Mobile IPv6 requires low-end obile devices to perfor all obility related signaling to aintain connectivity. Therefore, Internet Engineering Task Force IETF) proposed Proxy Mobile IPv6 PMIPv6) [2], a Network-based Localized Mobility Manageent NetLMM) solution which provides Internet connectivity to low power obile devices without requiring the to get involved in obility signaling. With the convergence of next generation wireless access technologies, such as 82.11, WiMAX, GPRS, 3G, etc, obile devices are now getting equipped with ultiple network interfaces that can facilitate increased availability, fault tolerance. This capability of counicating through ultiple network interfaces is soeties tered as ultihoing. Though PMIPv6 supports ultihoing, it has not been specified clearly how to achieve sealess handover in ultihoing scenario. Multi-hoed obile devices can aintain sealess connectivity with ultiple wireless access networks, thereby benefitting the ongoing sessions with its peers. We, therefore, proposed SEaless MObility using Shi6 SEMO6) [3], [4], a network-layer based solution that provides sealess connectivity to ulti-hoed obile devices. However, SEMO6 is a terinal-based protocol and it requires involveent of the low-power obile devices for obility anageent. A few works on PMIPv6 and related protocols have been reported in the literature. Iapichino et al. [5] propose a ulti-hoing solution cobining Host Identity Protocol with PMIPv6. Lee et al. [6] copares HMIPv6 and PMIPv6. Ki et al. [7] perfored handover analysis of PMIPv6. Iapichino et al. [8] perfored experiental evaluation of PMIPv6. However, none of the exiting works [6] [8] exploited the ultihoing feature of obile devices to iprove the perforance of PMIPv6. Although there exists a few entity-wise cost evaluation of obility protocols [9], [1] in the literature, no such evaluation for PMIPv6 has been attepted. Such an entity-wise evaluation is very crucial as obility anageent entities are very resource restricted and overloading of these entities ay result in coplete outage for the whole syste. Our objective of this work is to propose a network-based obility architecture that ensures sealess handover of obile devices through its ultihoing feature. We have proposed a novel obility architecture cobining the idea of PMIPv6 and SEMO6 [4] and naed it Proxy-SEMO6 that exploits ake-before-break strategy to achieve sealess handover in localized obility doain. In addition, we have perfored entity-wise cost evaluation of the proposed schee. The contributions of this work are: i) proposing a networkbased sealess obility solution for SEMO6, ii) developing an analytical odel to perfor signaling analysis of its key entities, and presenting nuerical results. Our results show the ipact of various network paraeters on the total overhead and perforance of the obility anageent entities of Proxy-SEMO6. Our proposed schee can guide design engineers to iprove the handoff perforance of network-based obility protocols that have received significant attention with the convergence of next generation all-ip networks. The rest of the paper is organized as follows. Section II explains basic SEMO6 [3] architecture in brief. In Section III, the proposed Proxy-SEMO6 architecture is explained along with its protocol operation and tiing diagra. Section IV presents a cost analysis of all the entities of Proxy-SEMO6, followed by the nuerical results in Section V. Finally, we conclude the paper in Section VI. II. SEMO6 ARCHITECTURE To achieve sealess handover of ultihoed obile devices, we earlier proposed SEMO6 [3], a terinal-based

2 Fig. 1. Proxy-SEMO6 architecture. obility solution in network layer. SEMO6 is based on Shi6 [11] and it decouples upper layer session identifier fro locator, thereby tries to reduce the ipact of obility on upper layer protocols, e.g., transport and application layers. SEMO6 focuses on providing a single platfor for obility and ultihoing. Moreover, it exploits ulti-hoing feature of Mobile Host MH) to iniize handover latency which is a very crucial factor for any handover protocol. However, as a terinal-based obility solution, the MH is responsible for all the obility-related signaling in SEMO6; thus, it consues significant aount of power a ajor concern for any obile device. Therefore, it is essential to design a network-based obility solution that relieves MH fro obility related signaling in addition to achieving sealess handover for ultihoed obile devices. III. PROXY-SEMO6 ARCHITECTURE Fig. 1 shows the Proxy-SEMO6 architecture which is siilar to Proxy MIPv6. The ain difference is the support of sealess obility for the ulti-hoed MH. Location Mobility Anchor LMA) keeps track of the MH in the localized obility doain and all traffic destined to the MH fro any Correspondent Node CN) are routed through the LMA. The Mobility Anchor Gateway MAG), usually the access router of the MH, perfors all obility related signaling on behalf of the MH, thereby saving MH s resources. In our proposed schee, we assued that LMA will allocate separate binding cache entry per interface of the MH. A. Tiing diagra The tiing diagra for the proposed Proxy-SEMO signaling is shown in Fig. 2. When any ultihoed MH enters the Proxy-SEMO6 doain, it attepts to attach to the access network L2 attachent) through one of its interfaces e.g., IF 1 ) and sends router solicitation to the access router e.g., MAG 1 ). Upon receiving the solicitation, the MAG uses MH s identification e.g., MAC address) to deterine whether the MH is authorized to use the localized obility support. The MAG then registers the MH s one interface IF 1 ) with the Fig. 2. Tiing diagra of Proxy-SEMO6 signaling. LMA through Proxy Binding Update PBU) which associates MAG 1 s address with MH s IF 1. The LMA adds a Binding Cache Entry BCE) into its table and establishes a bidirectional tunnel with the MAG 1 if one does not already exist. The LMA also sends Proxy Binding Acknowledgeent PBA) to the MAG 1 along with the assigned prefix. Upon receiving the PBA, the MAG 1 sends the MH Router Advertiseent that includes the allocated prefix), allowing MH to configure its IF 1 address through stateless auto-configuration. When the MH oves in a region that is covered by two access networks, it attepts to acquire another IP address fro the MAG 2 for its second interface IF 2 ) in a siilar way entioned above see Fig. 2). Thus, when the LMA receives the PBU fro the MAG 2, the LMA inserts another BCE into its table corresponding to the MH s IF 2 and establishes siilar bi-directional tunnel with the MAG 2. In addition, the LMA sends MH s CN a SEMO6 Update Request UR) essage to add this new IP address into CN s peer locator list. Thus, during the handoff, when the MH oves away fro the MAG 1 link, the L2 detachent) event is detected by the MAG 1 for exaple, through IPv6 neighbor unreachability detection event). The MAG 1 then infors the LMA about this event through the deregistration PBU which results in possible tunnel deletion if not needed for any other MH). At this point, the LMA sends CN UR essage so that CN can odify its peer locator list by reoving the MH s unreachable IP address. The CN can still continue the session using the unique ULID through the other locator corresponding to IF 2 ), thereby reducing handoff latency and packet loss.

3 B. Proxy-SEMO6 vs. PMIPv6 Proxy-SEMO6 attepts to reduce handover latency of PMIPv6 by exploiting the following strategies: Proxy-SEMO6 exploits the ake-before-break strategy, that is, attepts to connect through new MAG before disconnecting with the old MAG. It decouples the upper layer session identifier fro the location-based IP address, thereby ensuring session continuity, even though the MH s IP address varies due to the change of attachent point. However, Proxy-SEMO6 reduces handover latency of PMIPv6 at the cost of soe additional signaling e.g., sending UR essage to CN by the LMA). IV. SIGNALING COST ANALYSIS In this section, we perfor entity-wise cost analysis of Proxy-SEMO6. A. Assuptions and Notations To ake the odel analytically tractable, the following assuptions have been ade: Session arrival at each MH is equal. All session lengths are of equal size. Searching an entry in binding cache is assued to use binary search. Our cost analysis ignores standard IP switching costs. The notations used in the analysis are listed below. Nuber of MHs in the LMA-doain, N c Average nuber of CNs per MH, Nuber of MAGs in the LMA-doain, β M Unit transission cost for essage type M = {PBU, PBA, UR, UA,... } σ Proportionality constant of wireless link over wired link, ω Linear coefficient for lookup cost, T r Subnet residence tie, λ s Average session arrival rate, κ Maxiu transission unit, α Average session size, δ X Processing cost at entity X. B. Local Mobility Anchor In Proxy-SEMO6, the ajor costs for LMA are 1) initial registration cost, 2) binding update cost, 3) context establishent cost, 4) CN update cost, and 5) data delivery cost. 1) Initial registration cost: We assue that ɛ fraction of MHs enters into the LMA-doain fro outside at a rate of θ d. Thus, a total of ɛ θ d MHs enters into the LMA-doain per second. When a MH enters the LMA-doain, the MAG tracks it and sends PBU to the LMA. The LMA processes the PBU, allocates prefix for the MH, updates the binding cache and sends back PBA to the MAG. This incurs transission costs β PBU and β PBA ) and processing cost δ LMA ) on the LMA. Thus, the cost incurred at the LMA is, Λ Reg LMA = ɛnθ ) d βpbu +β PBA +δ LMA 1) 2) Binding update cost: When the MH oves within the LMA-doain and crosses subnets in every T r seconds), the concerned new) MAG detects it and sends the PBU to the LMA. In addition, MAGs send periodic refreshing updates to the LMA for extending the binding lifetie so that the entries are not reoved fro the cache. Let the binding lifetie bet l. Hence, frequency of sending periodic refreshing updates are η r = T r T /Tr l, and total frequency of sending LU and refreshing PBU is η t = 1 + T r ) T l /T r, This incurs transission cost β PBU and β PBA ) and lookup cost ωlog 2 ) at the LMA. Hence, ) Λ PBU LMA = η t βpbu +β PBA +ωlog 2 2) 3) Context establishent cost: The LMA establishes Shi6 context with the MH so that LMA can subsequently send update to CN about the locator change when MH oves to a new MAG-region). During the context establishent phase, the LMA is infored about the ULID, locator list and current locator used by the MH. Thus, it incurs transission cost for Shi6 Context β SC ) and processing by the LMA. Hence, Λ SC ) LMA = λ s 2βSC +δ LMA 3) 4) CN update cost: In order to odify the CN s peer locator list, LMA sends UR essage to CN whenever any new binding entry is inserted corresponding to MH s second interface. Furtherore, when MAG sends deregistration essage to the LMA, the LMA deletes corresponding binding entry and infors CN through UR essage. Thus, the cost on LMA regarding UR essage is as follows: Λ UR LMA = NNc ) βur +β UA 4) T r 5) Data delivery cost: In a session between the CN and MH, an average of α κ data packets and corresponding ACK) are transitted. The total data packet arrival rate to the LMA is λ p = N c λ s α κ. Data packets received by the LMA are tunneled through the respective MAG to the destination MH. This incurs transission cost for data and Ack packet β DP and β DA ) including extra IP-header β IP ), and lookup cost. Therefore, data delivery cost for the LMA is given ) by, Λ DD LMA = λ p βdp +β DA +β IP +ωlog 2 5) 6) Total Cost on LMA: Thus, the total cost on the LMA can be obtained by adding Eqns. 1), 2), 3), 4) and 5): Λ LMA = Λ Reg LMA +ΛPBU LMA +Λ UR LMA +Λ SC LMA +Λ DD LMA 6) C. Mobility Anchor Gateway The ajor costs for each MAG are 1) MH authentication cost, 2) binding update cost, and 3) data delivery cost. 1) MH authentication cost: For every registration request fro an incoing MH, the MAG is responsible for authentication. Based on the credentials AAA inforation) of the MH, the MAG either accepts or rejects the registration request. As a total of ɛ θ d MHs enters into the LMA-doain per second and if there are MAGs in the LMA-doain, therefore each MAG processes ɛ θ d / authentication requests. Hence, Λ AAA MAG = ɛnθ d δmag 7)

4 2) Binding update cost: Assuing MHs are uniforly distributed in the LMA-doain, each MAG sends PBU and refreshing PBU essages receives corresponding PBA) on behalf of / MHs. Moreover, the MAG has to detect the oveent of the MH inside its doain in to sense its arrival or departure, thereby requiring processing cost at MAG δ MAG ). Hence, cost incurred at each MAG for binding updates can be obtained as follows: Λ PBU MAG = ηtn ) βpbu +β PBA +δ MAG 8) 3) Data delivery cost: Each MAG receives data packets tunneled fro the LMA and decapsulate the and sends the to the corresponding MH. So the data delivery cost at each MAG is given by, α κ λ s βdp +β DA +β IP ) Λ DD MAG = NNc 9) 4) Total Cost on MAG: Thus, the total cost on each MAG can be obtained by adding Eqns. 7), 8) and 9): D. Mobile Host Λ MAG = Λ AAA MAG +Λ PBU MAG +Λ DD MAG 1) As a network-based obility solution, Proxy-SEMO6 incurs least cost at the MH. As all the encapsulation and decapsulation are perfored by the MAG for the packets to and fro the MH), the wireless access network is not affected by the extra IP headers. Other than packet delivery, the only task a MH is required to do is to establish Shi6 context with the LMA after registers with the LMA-doain. Thus, the total cost for each MH is as follows: α ) Λ MH = σn c λ s βdp +β DA +2λsβ SC 11) κ E. Efficiency We define a new etric called efficiency to easure the perforance of Proxy-SEMO6. It is defined as the ratio of net data delivery cost excluding all overheads) to the total cost that includes signaling and data delivery costs). The net data delivery cost of LMA can be expressed as follows: Λ Net DD ) LMA = λ p βdp +β DA 12) Hence, the efficiency of LMA is given by, ξ LMA = ΛNet DD LMA 13) Λ LMA The net data delivery cost of MAG can be expressed as follows: Λ Net DD MAG = NNc α ) s βdp +β DA 14) κ λ Hence, the efficiency of MAG is given by, ξ MAG = ΛNet DD MAG 15) Λ MAG The net data delivery cost of the MH can be expressed as follows: Λ Net DD α ) MH = σn c λ s βdp +β DA 16) κ Hence, the efficiency of MH is given by, ξ MH = ΛNet DD MH 17) Λ MH V. NUMERICAL RESULTS In this section, we present nuerical results showing ipact of various syste paraeters on Proxy-SEMO6 entities. The paraeter values used in nuerical analysis are derived using siilar approaches used in [9], [1]; each cost etric is a relative quantity and is based on the specific packet size unit cost for 1 bytes [9]). For exaple, β PBU =.76, since the size of PBU packet is 76. Therefore, we have set the paraeters as follows: β PBA =.76, β IP =.4, β DP = 5.72, β DA =.6, ɛ = 2%,θ d =.1, σ = 1, η =.3, η =.3, λ s =.1, δ LMA =.3, δ MAG =.3, κ = 512 bit/sec, α = 124 bits. The default values of other paraeters are = 1, T r = 7 sec, = 4, N c = 1. It can be noted that while coputing the total overhead of the LMA and MAG, we have considered all the factors considered including tunneling overhead, except payload delivery cost. A. LMA Fig. 3 shows the ipact of Session to Mobility Ratio SMR) on the total overhead of LMA for different nuber of MHs in the LMA-doain. SMR is the product of subnet residence tie T r ) and session arrival rate λ s ). We kept λ s =.1 fixed) while varying T r value. Again higher nuber of MHs under the LMA-doain increases the data delivery overhead on the LMA, resulting in higher overhead on the LMA. However, with respect to SMR, the total overhead on LMA decreases very slowly. This is due to the fact that higher SMR less obility rate) causes reduction in PBU cost, which is very sall copared to the data tunneling overhead. Hence the graphs are alost flat in nature. This is an interesting finding phenoena obtained fro our analysis. Fig. 4 shows the ipact of nuber of CNs per MH) on the total overhead of LMA. Total overhead increases for both increased nuber of CNs and higher session arrival rates. Higher session arrival affects data tunneling overhead since ore data packets are required to be tunneled through LMA. Fig. 5 shows the ipact of SMR on the efficiency of LMA. We find that efficiency increases for higher session lengths, having ore payload traffic copared to obility signaling traffic. In addition, the efficiency of LMA increases slowly with the increase of SMR since higher SMR less obility) causes reduced signaling traffic for the LMA. However, efficiency falls a little bit when the value of T r is 12 sec SMR = 1.2) and again in SMR = 2.4 due to rise in signaling traffic caused by refreshing updates. B. MAG Fig. 6 shows the ipact of subnet residence ties on the total overhead of each MAG, for different nuber of MHs in the PMIPv6-doain. The total overhead on each MAG decreases for higher values of T r, producing less PBUs. However, there are soe additional refreshing PBU sent to the LMA by the MAG, to keep the binding entry valid and those refreshing PBUs causes the overhead on MAG to rise in 12 sec and then 24 sec, which are ultiple of binding entry lifetie.

5 Total overhead on LMA N = 1 = 3 = Session to Mobility Ratio Total overhead on LMA 5 4 x 1 5 λ s =.1 λ =.4 s λ =.16 s Nuber of CNs 8 1 Efficiency of LMA %) α = 5k 52 α = 1k α = 2k Session to Mobility Ratio Fig. 3. Ipact of SMR on the total overhead of LMA for different nuber of MHs. Fig. 4. Ipact of nuber of CNs on total cost of LMA for different session arrival rates. Fig. 5. Ipact of SMR on the efficiency of LMA for different session lengths. Total Overhead on MAG = 1 = 3 = Subnet Residence Tie sec) Efficiency of MAG %) Nc = 1 Nc = 5 Nc = Session length bits) x 1 4 Signaling Overhed on MH %) α = 5kb α = 1kb α = 2kb Nuber of Correspondent Nodes Fig. 6. Ipact of subnet residence ties on total overhead of each MAG for different nuber of MHs in the PMIPv6-doain. Fig. 7. Ipact of session lengths on the efficiency of each MAG for different nuber of CNs. Fig. 8. Signaling overhead on the MH vs. nuber of CNs. Fig. 7 shows the ipact of session lengths on the efficiency of each MAG. Efficiency increases for larger sessions as net data delivery through the MAG increases. In addition, higher nuber of CNs also results in ore data traffic to the CNs, resulting in iproved efficiency. C. MH The percentage of signaling overhead on each MH is shown in Fig. 8. The only signaling responsibility of MH other than data transission cost) is to establish Shi6 context with the LMA after its registration with it. Results show that the overhead is very sall <.25%) for the MH; the higher the session length is, the lower the percentage overhead is, having ore payload traffic copared to signaling traffic. VI. CONCLUSION In this paper, we have proposed a novel sealess obility solution for network-based obility anageent that requires least involveent of the low-power obile devices. We have developed analytical odel to derive the total overhead on the obility anageent entities of the proposed schee. Results show interesting relationships aong various network paraeters, such as, network size, obility rate, traffic rate. Our proposed schee can guide design engineers to iprove the handoff perforance of network-based obility protocols that have received significant attention with the convergence of next generation all-ip networks. REFERENCES [1] C. Perkins, D. Johnson, and J. Arkko, Mobility support in IPv6, IETF RFC 6275, Jul 211. [2] S. Gundavelli and K. L. et al., Proxy Mobile IPv6, IETF RFC 5213, August 28. [3] M. S. Rahan and M. Atiquzzaan, SEMO6 - A ultihoing-based sealess obility anageent fraework, in IEEE MILCOM, San Diego, CA, Nov 17-19, 28. [4] M. S. Rahan, M. Atiquzzaan, W. Ivancic, and W. Eddy, Perforance Coparison between MIPv6 and SEMO6, in IEEE GLOBE- COM, Miai, FL, Dec 6-1, 21. [5] G. Iapichino and C. Bonnet, Host identity protocol and proxy obile IPv6: A secure global and localized obility anageent schee for ultihoed obile nodes, in IEEE GLOBECOM, Honolulu, HI, Nov 3- Dec 4, 29. [6] J.-H. Lee, Y.-H. Han, S. Gundavelli, and T.-M. Chung, A coparative perforance analysis on Hierarchical Mobile IPv6 and Proxy Mobile IPv6, Journal of Telecounication Systes, vol. 41, no. 4, pp , Aug. 29. [7] M. S. Ki, S. K. Lee, D. Cypher, and N. Golie, Fast handover latency analysis in Proxy Mobile IPv6, in IEEE Global Telecounications Conference, Miai, FL, Dec [8] G. Iapichino and C. Bonnet, Experiental evaluation of Proxy Mobile IPv6: An ipleentation perspective, in IEEE WCNC, Sydney, Australia, April [9] A. Shahriar, M. S. Hossain, and M. Atiquzzaan, A cost analysis fraework for NEMO prefix delegation-based schees, IEEE Transactions on Mobile Coputing, vol. 11, no. 2, pp , Jul 212. [1] M. S. Hossain, M. Atiquzzaan, and W. Ivancic, Cost analysis of obility anageent entities of SINEMO, in IEEE ICC, Kyoto, Japan, Jun 5-9, 211. [11] E. Nordark and M. Bagnulo, Shi6: Level 3 ultihoing shi protocol for IPv6, IETF RFC 5533, Jun 29.