Review of Measurements, Modelling, and Analysis of a P2P System (by Gummadi et al)

From: Prasanna Kumar Jayapal (prasak_at_winse.microsoft.com)
Date: Mon Mar 08 2004 - 17:39:29 PST

  • Next message: Slavik Krassovsky: "Krishna Gummadi @ al Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload."

    This paper ("Measurement, Modeling, and Analysis of a P2P workload")
    analyses the file-sharing workload characteristics of a p2p system and
    compares it to the typical web surfing characteristics. The authors
    present a nice report of their systematic analysis and show that the
    fetch-at-most-once behavior (due to the immutable nature of the p2p
    objects) of the p2p system causes the major difference with the web
    workload. The key point made in this paper is that the distributed data
    caching will be effective in reducing the bandwidth consumption of the
    p2p system.

    The data for this analysis is based on a 200-day trace of Kazaa p2p
    traffic collected at UW and the methodology is nicely described in the
    paper. However, Kazaa is just one of the numerous p2p systems available
    and I would have liked to see some observations on other systems as well
    since the paper targets p2p systems in general. The main qualities of a
    p2p system (or Kazaa) are - p2p users are more patient, user queries
    reduce as they age, workload consists of a large set of immutable
    objects, users fetch the objects at most once.

    A model was created based on the statistics observed which showed that
    p2p file-sharing systems are driven primarily by introduction of new
    data objects and new clients joining the system and also by clients'
    fetch-at-most-once behaviors. These behaviors differ from Zipf curves
    which are often used to describe web behaviors. I was not very familiar
    with the Zipf curves and I felt a little difficult to follow them. From
    the analysis of the model it was concluded that, without the
    introduction of new objects and clients, p2p system performance
    decreases over time and the new objects act as a rejuvenating factor
    that counter-balances the impact of fetch-at-most-once behavior.

    The authors then used their workload to explore the exploitation of
    locality and caching in the p2p systems to reduce the network traffic.
    They discuss different techniques like organizational proxy cache,
    request redirectors and locality-aware mechanism and finally demonstrate
    that the locality aware mechanism would reduce the external bandwidth
    consumption by maximizing the use of data stored in the local peers. In
    summary, there is a good deal of locality in the p2p system where a
    protocol for caching or redirecting can potentially lighten the
    bandwidth load for the entire network. Overall, this was an interesting
    paper to read and very insightful.


  • Next message: Slavik Krassovsky: "Krishna Gummadi @ al Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload."

    This archive was generated by hypermail 2.1.6 : Mon Mar 08 2004 - 17:39:17 PST