Context Caching using Neighbor Graphs for Fast Handoffs in a Wireless Network Arunesh Mishra Min-ho Shin William A. Arbaugh Computer Science Department University of Maryland Publication：INFOCOM Date ：March 2004 On page ：351 ~ 361 Reference ：19 by Xinchang 2007/1/12

Outline Introduction Related Work Context Transfer protocol IAPP Handoff Concepts Neighbor Graph Proactive Caching Algorithm Performance Analysis Experiment Simulation Conclusion Roam 即為 paper 所說的 discrete mobility

Introduction Wireless Network mobility： discrete mobility (like roam) continuous mobility (like seamless handoff) Voice and Multimedia are usual in continuous mobility Total latency(L2 + L3) of handoff must be fast (< 50 ms) Unfortunately ,the most Wi-Fi network Not achieve this Goal !! How to reduce handoff latency ?? proactive context caching mechanism base on NG Only Layer 2 latency approximately to 90% of all, exceeds 100ms Latency of Handovers the time between : the last moment where the MN can receive and send packets through the old AR the first moment where MN can receive and send packets through the new AR. This Paper want to improve reassociation latency

Relate Work Context Transfer Topology algorithm R. Koodli and C.E. Perkins, “Fast Handover and Context Relocation in Mobile Networks” M. Nakhjiri, C. Perkins, and R. Koodli, “Context Transfer Protocol,” Topology algorithm Sangheon Pack and Yanghee Choi. “Fast Inter-AP Handoff using Predictive-Authentication Scheme in a Public Wireless LAN,” “Pre-Authenticated Fast Handoff in a Public Wireless LAN based on IEEE 802.1x Model,” S. Capkun, Levente Buttyan, and Jean-Pierre Hubaux, “Self-Organized Public-Key Management for MANET” Radia Perlman, “An algorithm for distributed computation of a spanningtreein an extended lan,”

Context Transfer protocol Context Transfer protocol (CXTP) enable authorized context transfer Why need to perform Context transfer between APs ?? support Node mobility can operate with minimal disruption provide an interoperable solution to support L2 radio access Context Transfer takes place when an event Trigger pAR transfer the context nAR request the context MN send a message to AR to transfer context What informations of Context ?? MN’s IP address QoS , security , header compression,AAA information Minimal disruption 即想要做到 seamless mobility 支援 IPv4 和 IPv6 在reassociation 時，the APs involved exchange station context message using IAPP between APs between AP and the RAUDIO server

IAPP increase overall handoff latency of their reactive fashion IAPP play a significant role during handoff Single Association Invariant Secure transfer of state and context information between APs Two types of interaction for Complete Context transfer between APs between AP and the RADIUS server Two message will trigger IAPP Association Reassociation IAPP increase overall handoff latency of their reactive fashion

Handoff Concepts Handoff Types Handoff Scenarios Horizontal Handoff The handoff in homogeneous networks Intrasystem Handoff Vertical Handoff The handoff between heterogeneous networks Intersystem Handoff Handoff Scenarios L2 handoff MN’s movement between APs belong to a common IP’subset L3 handoff MN’s movement between APs belong to a different domain

Handoff Scenarios

Handoff Procedure Handoff Procedure 1.discovery phase 2.reauthentication phase or 1.scan phase 2.authentication phase 3.reassociation phase [Note] 講一下此圖為何為 reactive fashion?? 說明 neighbor graph => proactive fashion

Neighbor Graph Definition Reassociation Relation Two APs, said APi and APj , are said to have reassociation relationship if it is possible for an STA to perform an reassociation through some path of motion between locationsof APi and APj Association Pattern Γ(c) for a client c is association pattern as {(ap1, t1), (ap2, t2), . . . , (apn, tn)} where api is the AP which the client reassociates at time ti AP Neighbor Graph Define a undirected graph G = (V , E) where V = {ap1, ap2, . . . , apn} is the set of all APs E = { e / e =(api, apj) if between api, apj have reassociation relation }

Neighbor Graph Neighbor(api) = {apik : apik ∈ V, (api, apik ) ∈ E}

Neighbor Graph Generation of NG automatically generate by individual APs over times two ways that APs can learn the edges in NG AP receive 802.11 reassociate request from STA AP receive Move-Notify from another AP via IAPP Each AP maintains edges locally (use LRU) eliminate the outlier [Situation] client that goes into power saving mode, and wake up in a different location to reassociation to any other AP Timestamp base LRU would guarantee freshness of NG over time

Proactive Caching Algorithm Proactive Caching & Locality of Mobility Proactive Caching Algorithm Function/Notation Context(c) Cach(apk) Propagate_Context(api, c , apj) Obtain_Context(apfrom, c , apto) Remove_Context(apold, c , apnghbr) Insert_Cache(apj , Context(c)) Who will perform this Algorithm ?? Cache replacement mechanism LRU

Proactive Caching Algorithm

Performance Analysis Upper bound on the cache size Clientlist(api) = the set of clients associated to api Napi = | Neighbor(api) | = degree(api) M = maximum number of clients associated to any AP

Performance Analysis Characterizing the Cache Misses we expect 100% cache hit ratio for reassociation (before) if following is hold： NG has been learned edge Cache size is unlimited Two kinds of cache miss possible during a reassociation Reassociation between non-neighbor APs Context evicted by LRU replacement This mean Cache at each AP is large enough according to eq (1) A faster client spent less time at AP, have higher probablity of cache hit

Performance Analysis IAPP and Proactive Caching Cache-Notify Cache-Response Cache-Invalidate

Experiment Measure 114 reassociations in the Testbed resulting average reassociations latency Wireless Testbed five AP at 3rd floor four AP at 2nd floor IAPP, neighbor graph, caching algorithm implement in driver 實驗: 實作 IAPP with Neighbor Graph

Experiment Experiment process Experiment Result Mobile unit Two experiments Experiment A Experiment B Experiment Result cache-miss without an outlier：15.37ms cache-miss with the outlier：23.58ms cache-hit：1.7ms

Experiment Experiment A

Simulation Simulation Objective Simulation Model and Assumptions To observe the effect of cache size, the number of clients, the mobility of clients on the cache hit ratio To observe the performance of caching with various NGs Simulation Model and Assumptions AP Neighbor Graph does not change during the simulation Correctness and completeness of the Neighbor Graphs Initial User-AP distribution Roaming Model User Mobility roaming model => (a) (b) (c) user mobility => mobility index

Simulation Simulation Environment Simulation Result random and connected NG with 10, 20, 50 and 100 vertices Duration of the Simulation The simulation runs for one million reassociation events Simulation Result Mobility Improves Proactive Caching Performance Effect of Cache Size and Client Mobility on Hit Ratio Effect of Cache Size and Number of Users on Hit Ratio 15 % 的 cache size 就可以讓 hit ratio 達到 98%

Simulation

Simulation

Simulation In Fig. 11, We find 15% Cache Size is sufficient for a hit ratio of 98%

Conclusion Introduce a novel, efficient and dynamic data structure, Neighbor Graph(NG) NG, which capture the topology of a wireless network by automously monitoring Network NG can also used to eliminate the expensive scanning operation for faster MAC handoff Implement the proactive caching algorithm base on NG for faster wireless handoff In Simulation, find that cache size plays an important role in the performance of algorithm