在行動隨意網路中利用區域廣播來協助快取資料分享

1. 在行動隨意網路中利用區域廣播來協助快取資料分享在行動隨意網路中利用區域廣播來協助快取資料分享 Exploiting In-Zone Broadcasts for Cache Sharing in Mobile Ad Hoc Networks IEEE TRANSACTIONS ON MOBILE COMPUTING, 2009 演講者: 楊政儒 副教授 　　　　　　　　　 九十八年十月三十日

2. Outline • Introduction • Caching Sharing Techniques in MANETs • Push-Based Cache Sharing Scheme (IXP) • Pull-Based Caching Sharing Scheme (DPIP) • Performance Evaluation • Conclusion • Future work

3. Introduction • Mobile Ad hoc Network (MANET) • Mobile nodes form a dynamic and temporary communication network • Each node acts like a router to forward packets for others • Without any pre-existing infrastructure • Network topology changes frequently • Extensive research in ad-hoc network focus on network-layer protocol • Proactive routing protocol (DSDV) • Reactive routing protocol (AODV) • Hybrid (ZRP) • The research focus on the issue of data access in a MANET

4. Introduction • Data Access in a MANET Data Server Ad hoc network Requester

5. Introduction • Problems • Having mobile devices always retrieve data from the data server may result in a large amount of traffic • Mobile host may suffer from high access latency • Increase packet loss probability for long-distance data access • Traffic near the data server will be heavy, and this leads to a potential performance bottleneck • Goal • Avoid above problems by exploiting cache sharing techniques

6. Cache Sharing Techniques in MANETs • Cache Sharing • Allows neighboring mobile nodes to access each other’s cache contents • The number of long-distance data accesses to the data server can be reduced • The Key to Cache Sharing • A node has to know if there is some node in its vicinity that has cached the data it requires and where it is, if any • Cache replacement police

7. Cache Sharing Techniques in MANETs • Two types of caching sharing techniques • Push-based • Pull-based (describe later) • Push-based scheme • When a node caches a new data item, it actively advertises the caching event to the nodes in its neighborhood • Mobile nodes in the vicinity will record the caching information upon receiving such an advertisement

8. Push-Based scheme IXP • Concept of IXP • Having each node share its cache contents with the nodes in its zone • Whenever a node M caches a data item, it broadcasts an index packet to its buddies to advertise the caching event The nodes in zone of M • Each node maintains an index vector IV to record caching information

9. Push-Based scheme IXP • Index Vector (IV) • N elements (N is the number of data items) • Each element contains 3 entries to record caching information • IV[x].cached • Record whether x is cached locally • IV[x].cachednode • Record a nearby node that has cached x • IV[x].count • Contains a count of M’s buddies that are known to have cached x since x is cached

10. Push-Based scheme IXP IV[y].cachenode=MNULL Broadcast an index packet <M, x,y> IV[x].cachednode=M M cache x and replace y <DATA X> <M,x,y> <M,x,y> <M,x,y> <M,x,y> M1 <M,x,y> M2 Want to access x M’s zone <request> <request> 10

11. Push-Based scheme IXP • Cache Replacement • Count-based scheme (CV) • CV scheme • M resets the IV[x].count to 0 when it first caches a data item x • When M receive index packet <buddy,x,y> • IV[x].count ++ , IV[y].count -- • M replaces the data item with maximum value of IV[].count among all cached ones

12. Push-Based scheme IXP • Looping Problem • Node mobility in a MANET may cause caching information to become misleading

13. Push-Based scheme IXP • Solution of Looping Problem • Make the request packet that has been received by any intermediary node twice to be sent to the data server without further detouring

14. Pull-Based scheme DPIP • Drawback of IXP scheme • Index Vector (IV) may become obsolete. • Example Pull-based approach overcomes this problem

15. Pull-Based scheme DPIP • Pull-Based Approach • When a mobile node wants to access a data item that is not cached locally it will broadcast a request to the nodes in its vicinity • A nearby node that has cached the data will send a copy of the data to the request originator (a pull operation)

16. Pull-Based scheme DPIP • Drawbacks of Pull-Based Approach • Extra access latency • Need to wait for a timeout interval • Pulling effort may be in vain • Extra communication overhead • More than one copy may be returned to the request originator • Goal • Propose a sophisticated pull-based protocol to deal with above problems

17. Pull-Based scheme DPIP • DPIP Functions • Data pulling • Index pushing (in implicit manner)

18. Pull-Based scheme DPIP M wants to access x and will replace y after caching x. M broadcasts a dp packet <M,x,y> to its buddies IV[x].cachednode=M Implicit pushing (index push) M2 <M,x,y> <M,x,y> <M,x,y> <M,x,y> M3 cached x and will return x to M M3 M4 Data pulling

19. Pull-Based scheme DPIP • Properties of DPIP • The in-zone dp broadcast allow DPIP to use the latest cache contents • DPIP induces less in-zone broadcasts than IXP. • If M’s IV[x].cachednode is not NULL, M need not broadcast dp for requesting x • The scope of cache sharing is extended by a factor of two

20. Pull-Based scheme DPIP Extended scope for data sharing

21. Pull-Based scheme DPIP • DPIP overcomes the drawbacks of pull-based approach • Extra access latency • Extra communication overhead • Pulling effort may be in vain M can send its request to the buddy node recorded in IV[].cachenode Provide wider range of data sharing

22. Performance Evaluation • Simulation Setup • Ns-2 with CMU wireless extension. • Nodes move in an area of 1500mx500m • Random way point movement model is used • One server is placed at the upper left-hand corner of the rectangle area • Wireless transmission range is 250m • Channel capacity is 2Mbps • Speed of mobile node is selected randomly from 2 to 20m/s • The number of mobile nodes is 70 • The radius of zone is set to 1

23. Performance Evaluation • Compare our protocol with HybridCache (IEEE Transactions on Mobile Computing, 2006) • The scope of caching sharing is only on the forwarding path of a request to the data server. Data X Record R1 cached X Request X Data X Request X

24. Performance Evaluation • Performance Measure • Cache ratio (for IXP and DPIP) • local_hit • zone_hit_path (acquire data via IV[].cachednode) • zone_hit_broadcast (acquire data via dp broadcast) • zone_hit_location (acquire data via location_reply packet) • Request failure ratio • Access time • Network traffic introduced by a data request • Ratio of request served by the data server For DPIP 24

25. Simulation Results (Effects of Cache Size)

26. Simulation Results (Effects of Cache Size)

27. Simulation Results (Effects of Cache Size)

28. Simulation Results (Effects of Cache Size)

29. Simulation Results (Effects of Cache Size)

30. Simulation Results (Effects of Cache Size)

31. Simulation Results (Effects of Node Mobility)

32. Non-uniform Data Access using Different Cache Replacement Policies SXO is a LFU-like scheme • Zipf-like distribution is used to model data access

33. Simulation Results ( Data of Different Sizes) SXC=Size*Count SXC=CV If all data are of the same size

34. Conclusion • Propose two protocols, IXP and DPIP • Fully exploit in-zone broadcasts to facilitate cache sharing operation in a MANET • Design a simple but efficient count-based cache replacement scheme • IXP uses explicit index broadcasts, while DPIP offers implicit index push • DPIP provides a wider cache sharing range than IXP

35. Future work • Pull with Piggybacked Push scheme • Overcome the drawbacks of DPIP • Implicit index push may be incorrect • Incorrect/Inaccurate caching information may lead to request failure or detour

36. Future work • Optimal Storage Placement for Tree-Structured Networks • Given a set of source nodes generate data and forward to some halfway storage nodes • The requesting node retrieve needed data from these storage nodes • Goal Determine an optimal set of storage nodes that minimizes overall communication cost