P2P Kazaa Review

From: ahemavathy (ahemavathy_at_hotmail.com)
Date: Sat Mar 06 2004 - 20:48:23 PST

  • Next message: Tarik Nesh-Nash: ""Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload" Review"

    Peer to Peer file sharing is very popular these days and their traffic consumes a large Internet bandwidth.The fundamental property of such systems is very different from other web applications primarily in 1) the objects are immutable and hence fetched at most once 2) primary forces in the system are creation of new objects and the addition of new users. Also the expectations and behavior of users of such systems are different from other web applications: for one they are very patient :) and over a period of time their demand decreases

    The authors study Kazaa over a period of 6 months and come to interesting conclusions about P2P file sharing systems. The main conclusion is popularity distribution of Kazaa is not Zipf since clients fetch objects at most once. Correspondingly there is a decrease in the hit rate over time. New objects will improve the performance but it is not the same for new clients as eventually the constant arrival rate cannot compensate for the increasing no. of old clients. Finally the authors show how locality awareness will significantly improve the hit rate for file sharing systems. That can be done by using a proxy cache or organization based locality aware mechanisms for reducing external downloads (centralized redirector or decentralized supernodes). Their measurements show that even with pessimistic availability the improvement in hit rate is very good. I found the paper very interesting but not convinced if the suggested improvements are feasible due to the legal questions of these systems. But surely P2P file sharing will be a success performance wise if it is made locality aware. But if that were to happen then P2P sharing will have to be legal and might no longer be as popular !


  • Next message: Tarik Nesh-Nash: ""Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload" Review"

    This archive was generated by hypermail 2.1.6 : Sat Mar 06 2004 - 20:49:42 PST