Content Delivery review

From: Kevin Wampler (wampler@cs.washington.edu)
Date: Mon Nov 08 2004 - 02:38:03 PST

  • Next message: Pravin Bhat: "(no subject)"

    In "An Analysis of Internet Content Delivery Systems" the authors analyze
    the characteristics of TCP traffic from the University of Washington over
    nine days and categorize it into the traffic into that is generated from
    WWW, Akamal, and P2P usage. They find that the lion's share of the
    traffic is due to P2P transfers, that the average file transferred over
    P2P networks is a thousand-fold larger than the average WWW file, and that
    a few P2P users account for the bulk of P2P traffic. The analysis is
    clear, but thorough, highlighting the important differences between the
    traffic types. The discussion about P2P caching is particularly useful,
    as it provides a way to cut down on the large impact of P2P traffic
    without simply disallowing it or capping its usage.

    This paper also calls into question the scalability of P2P file sharing
    systems, both in terms of the massive bandwidth they demand and in terms
    of the noted high usage of a few supernodes in the KaZaA network. As the
    file transfers over P2P represent what is very likely to become a
    commonplace occurrence as the availability of legal multimedia increases,
    capping the usage of such things does seem to be a temporary fix at best.
    The caching scheme proposed by the authors looks to be much more
    promising, but it is limited to caches built up at centralized locations
    (such as at a university gateway). It would be interesting to see if
    these same results could be achieved with some reasonable amount of
    caching in the modes (or supernodes) of a P2P network. This decentralized
    approach has the advantage that it would apply to P2P usage across the
    internet, rather than just in the University setting. BitTorret can be
    seen as partially addressing this issue, but it seems possible that files
    could be transfered through (and cached at) intermediate nodes in a P2P
    transfer and that by doing so the load of the traffic could be even better
    managed.


  • Next message: Pravin Bhat: "(no subject)"

    This archive was generated by hypermail 2.1.6 : Mon Nov 08 2004 - 02:38:04 PST