A Novel Approach for Cooperative Overlay-Maintenance in - PowerPoint PPT Presentation

A Novel Approach for Cooperative Overlay-Maintenance in Multi-Overlay Environments 1 Wu-Chun Chung, National Tsing Hua University 2010/11/30

A Novel Approach for Cooperative Overlay-Maintenance in Multi-Overlay Environments Chin-Jung Hsu, CS, National Tsing Hua University, Taiwan Wu-Chun Chung, CS, National Tsing Hua University, Taiwan Kuan-Chou Lai, CIS, National Taichung University, Taiwan Kuan-Ching Li, CSIE, Providence University, Taiwan Yeh-Ching Chung, CS, National Tsing Hua University, Taiwan 2 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Outline  Introduction  Related Work  Cooperative Strategy  CFD – failure detection  CNPE – network-proximity estimation  Experimental Results  Conclusions 3 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Introduction  Overlay Network A virtual network overlay another layer  Chord, Gnutella, Super-Peer model, etc.  Focus: over the Internet Name IP Port User A 66.238.93.162 80 User B 220.74.26.128 823 User C 118.169.74.72 8080 User D 66.238.93.162 168 140.114.91.88 . 220.74.26.168 . . 66.238.93.162 209.131.36.158 118.169.74.72 4 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Internet Visualization Wikipedia: Opte Project 5 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Example: Ring-based overlay Wikipedia: Opte Project 6 Wu-Chun Chung, National Tsing Hua University 2010/11/30

More and more applications  Overlay-based applications are growing  P2P file sharing – gnutella, eDonkey, BitTorrent, etc.  P2P Steaming – PPStream, PPLive, Joost, etc.  Resource Discovery – Mercury, MAAN, etc.  Cloud computing – Cassandra, Hadoop, etc.  Multiple overlays co-habit the Internet 7 Wu-Chun Chung, National Tsing Hua University 2010/11/30

A multi-overlay environment (MOE) 8 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Each overlay network may serve an application 140.114.91.88 220.74.26.168 66.238.93.162 209.131.36.158 118.169.74.72 9 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Motivation  Overlay network introduces maintenance cost  failure detection  latency/bandwidth measurement  routing table adjustment  adaptive approach  … etc.  n * Cost = large  how to reduce?  Some of these overlay-maintenance costs are redundant 10 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Related Work  [2007][ICDCS] Build One, Get One Free: Leveraging the Coexistence of Multiple P2P Overlay Networks  Sharing information to reduce maintenance cost  Focus on two specific overlays  [2009][DAIS] Exploiting Synergies between Coexisting Overlays  A comprehensive consideration on the reduction of maintenance costs  Lack of the consideration of intersection ratio 11 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Intersection Ratio  the percentage of nodes which locates in both overlays 25% Overlay A Overlay B Intersection Ratio 12 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Objective  Multi-overlay environments  Reduce the total maintenance cost  Propose a general approach  Consider a realistic MOE environment 13 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Cooperative Strategy  To reduce the maintenance costs  The total cost could be smaller Overlay A Overlay B Overlay C n 1 n 2 n 3 Cooperative Maintenance n < n 1 +n 2 +n 3 14 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Master-Slave Model  One overlay is selected to be the master  The master overlay could help reduce the common maintenance operations Master Overlay A Slave Overlay B Overlay C Slave n 1 n 2 n 3 Cooperative Maintenance n < n 1 +n 2 +n 3 15 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Master-Slave Model  Two kinds of inter-overlay protocols to support two types of overlay maintenance Periodical State Maintenance Sharing 16 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Inter-Overlay Protocols  Subscribe/Notify protocol   periodical maintenance  E.g. failure detection  periodically checks the status of neighbor nodes to ensure the routing mechanism  Query/Response protocol   state sharing  E.g. network-proximity estimation  share the information of network state to make the decision of routing path 17 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Cooperative Failure Detection (CFD) master slave A A probe probe B B Elimination 18 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Cooperative Failure Detection (CFD) master slave A C A probe probe B B Cooperation 19 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CFD – Subscription Process 1. subscribe probe inform A B A B probe forward E E C D C D 2. notify master slave 20 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CFD – Notification Process 1. subscribe A B A B notify probe E E C D C D 2. notify master slave 21 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Cooperative Network-Proximity Estimation (CNPE) master slave A A D C B ? ? ? Elimination 22 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Cooperative Network-Proximity Estimation (CNPE) master slave A A D C B ? ? ? ? ? Exploration E F 23 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CNPE – Query/Response Process I 1. query A B A B latency: <50ms(2) 30ms 20ms 45ms E E 35ms E (30ms), C (45ms) C D C D 2. response master slave 24 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CNPE – Query/Response Process II 1. query Neighbor’s Neighbor A B A B latency: <50ms(3) 30ms 20ms 45ms E E E (30ms), C (45ms), 35ms B (50ms) C D C D 2. response master slave 25 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Experimental Environment  PeerSim simulator  Cycle-based simulation engine  Unstructured, Ring, Tree Overlays  Parameter K: neighbor numbers  Comparison metric: reduction rate 26 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Comparison Metric –Reduction Ratio  The higher the reduction ratio is, the more efficient our approach will be CFD CNPE 27 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Experimental Results –Session Time CFD CNPE 50.0% 70.0% 45.0% 60.0% 40.0% 50.0% 35.0% Reduction rate Reduction rate 30.0% 40.0% 25.0% 30.0% Master: Unstructured(K=4), Slave: Ring(K=2) Master: Proximity(K=4, N=2), Slave: Proximity(K=6, N=2) 20.0% Master: Unstructured(K=4), Slave: Tree(K=3) 15.0% 20.0% Master: Ring(K=2), Slave: Unstructured(K=4) Master: Proximity(K=6, N=2), Slave: Proximity(K=4, N=2) Master: Ring(K=2), Slave: Tree(K=3) 10.0% 10.0% Master: Tree(K=3), Slave: Unstructured(K=4) 5.0% Master: Tree(K=3), Slave: Ring(K=2) 0.0% 0.0% 0 20 40 60 80 100 120 0.0 20.0 40.0 60.0 80.0 100.0 120.0 Median session time (min.) Median session time (min.) 28 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Experimental Results –Intersection Ratio CFD CNPE 50% 50% Master: Proximity(K=4, N=2), Slave: Proximity(K=6, N=2) Master: Unstructured(K=4), Slave: Unstructured(K=6) 45% Master: Proximity(K=6, N=2), Slave: Proximity(K=4, N=2) Master: Unstructured(K=6), Slave: Unstructured(K=4) 40% 40% 35% Reduction rate Reduction rate 30% 30% 25% 20% 20% 10% 15% 10% 0% 5% 0% 25% 50% 75% 100% -10% 0% Intersection ratio 0% 25% 50% 75% 100% Intersection ratio 29 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CFD + CNPE Overlay C Overlay B Ring (K=2) Unstructured (K=6) 50% 25% Overlay A Unstructured (K=4) 75% Master Master Overlay D CFD CNPE Tree (K=3) 30 Wu-Chun Chung, National Tsing Hua University 2010/11/30

CFD + CNPE  The total reduction rate approximates 40% 0.45 39.42% 0.4 0.35 Reduction ratio 0.3 24.88% 0.25 0.2 14.54% 0.15 0.1 0.05 0 CFD CNPE Hybrid 31 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Conclusions  Multi-overlay environments have emerged  Total maintenance cost is high  Some operations are redundant  Cooperative maintenance approach  A general Master-Slave model CFD – Subscribe/Notify protocol 1) CNPE – Query/Response protocol 2)  Reduce more than 60% 32 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Conclusions Maintain one, Get many free 33 Wu-Chun Chung, National Tsing Hua University 2010/11/30

Future Work  Other operations of overlay maintenance  Master overlay selection criteria  Automatic selection mechanism 34 Wu-Chun Chung, National Tsing Hua University 2010/11/30

A Novel Approach for Cooperative Overlay-Maintenance in Multi- Overlay Environments 35 Wu-Chun Chung, National Tsing Hua University 2010/11/30