Docstoc

ppt - Improving Unstructured Peer-to-Peer Systems by Adaptive

Document Sample
ppt - Improving Unstructured Peer-to-Peer Systems by Adaptive Powered By Docstoc
					 Tsung-Han Lin 9/21/05




Improving Unstructured Peer-to-Peer Systems
   by Adaptive Connection Establishment




Found in:IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 9,
SEPTEMBER 2005
Author:Li Xiao, Member, IEEE, Yunhao Liu, Member, IEEE, and Lionel M.
Ni, Fellow, IEEE
Introduction

   The objective of this paper is to minimize the
    effect due to topology mismatch
   This paper propose the Adaptive Connection
    Establishment (ACE) that builds an overlay
    multicast tree among each source node and
    its neighbor peers
   ACE is scalable and completely distributed, it
    also shows that ACE can improve
    performance
Related work

   To improve search performance and reduce unnecessary
    flooding traffic in unstructured P2P systems
    –   Route queries to peers that are likely to have the requested items.
    –   Index cache
            Both approaches are limited by the topology mismatch problem
    –   To cluster close peers based on their IP addresses
            Mapping accuracy is not guaranteed
            May affect the search scope in P2P networks
    –   Location-aware topology match scheme
            More overhead and requires that the clocks in all peers be
             synchronized
Adaptive Connection Establishment
-Unnecessary Message Duplication

   Unnecessary Traffic by Flooding
Adaptive Connection Establishment
-Unnecessary Message Duplication

   Topology Mismatch Problem
Adaptive Connection Establishment
-Unnecessary Message Duplication

   Topology Mismatch Problem
    –   More than 70% of paths suffer from this problem
Adaptive Connection Establishment


   Average traffic cost
    –   C=M x L
            M: the number of messages that traverse the overlay
             connection
            L: the number of physical links in this overlay connection
   Query response time
    –   the time period from when the query is issued until when the
        source peer received a response result from the first
        responder.
Adaptive Connection Establishment
-Design of ACE

   Phase 1: Neighbor Cost Table Construction and
    Exchanging
    –   Each peer probes the costs with its immediate logical
        neighbors and forms a neighbor cost table
            Cost: network delay between two peering nodes
    –   Two neighboring peers exchange their neighbor cost tables
            so that a peer can obtain the cost between any pair of its
             logical neighbors.
   How often peers exchange their neighbors’ cost
    table
    –   Periodic approach
    –   Even-driven approach
Adaptive Connection Establishment
-Design of ACE

   Phase 2: Selective Flooding
    –   Based on neighbor cost tables, MST (among each peer S and its
        immediate logical neighbors)
    –   Flooding/nonflooding-neighbors
Adaptive Connection Establishment
-Design of ACE

   Phase 3: Overlay Optimization
    –   A peer tries to replace those physically far away
        neighbors by physically close by neighbors, thus
        minimizing the topology mismatch traffic
Adaptive Connection Establishment
-Design of ACE
                             SH<SG

   Phase 3:
                               SH<SG




               SH>SG,GH>SH    SH>SG,GH<SH
Adaptive Connection Establishment
-Depth of Optimization

   Define h-neighbor closure of a source peer
    as the set of peers within h hops from the
    source peer.
    –   A larger value of h leads to a better topology
        matching improvement, but a higher overhead
        due to the extra information exchanging.
Adaptive Connection Establishment
-Depth of Optimization


                  1-Neighbor closure
Adaptive Connection Establishment
-Depth of Optimization
Simulation

   Performance Metrics
    –   Traffic cost
    –   Response time
    –   Search scope
    –   Dynamic P2P environment
Simulation

   ACE in Static Environments




                                 65% off
          60%~70% off
Simulation
   ACE in Static Environments




                                                  35% off


   Even if we vary the size of the overlay topology ranging from
    1000 to 9000, and the results are consistent
Simulation

   ACE in Dynamic Environment
    –   Assume:
            Peer avg. lifetime is 10 minutes
            Every minute each peer issued 0.3 queries
            9000 peers
            Avg. neighbors: 6
Simulation

   ACE in Dynamic Environment
Simulation

   ACE in Dynamic Environment
Simulation

   The Impact of Optimization Depth
    –   E: avg. number of neighbors
Conclusion

   ACE will make the decentralized flooding-
    based P2P file sharing systems more
    scalable and efficient

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:84
posted:5/15/2011
language:English
pages:22