Review of "An Analysis of Internet Content Delivery Systems"

From: Michelle Liu (liujing@u.washington.edu)
Date: Sun Nov 07 2004 - 16:03:16 PST

  • Next message: Tom Christiansen: "Saroiu et al, 2002"

    Review of "An Analysis of Internet Content Delivery Systems"

    Jing Liu

     

        This paper presents an analysis of Internet content delivery systems, including HTTP World-Wide Web (WWW), content delivery networks (CDNs) and peer-to-peer file sharing systems (P2P).

        The author clearly states the methodology of analysis. Passive network monitoring was used to collect traces of traffic flowing between the University of Washington (UW) and the rest of the Internet. The traffics are separated into types of WWW, Akamai (CDN system), P2P (the union of Gnutella and Kazaa) and non-HTTP TCP traffic. The detailed content delivery characteristics are described with respect to objects, clients, servers and scalability of peer-to-peer systems. Several important observations are obtained. First, the object median size of P2P traffic is three orders of magnitude larger than that of web objects. Second, a small number of P2P users are consuming a high fraction of bandwidth. Third, while the P2P request rate is quite low, the transfers last long, resuling in many simultaneous P2P connections. Fourth, while the design of P2P overlay structures focuses on spreading the workload for scalability, a small number of servers are taking the majority of the burden.

        Caching is also studies in this paper. Widely-deployed proxy caches would significantly reduce the need for a separate content delivery network. Moreover, caching would have a large effect on a wide-scale P2P system, potentially reducing the wide-area bandwidth demands dramatically.

        Since the internet traffic has experienced a dramatic shift from WWW traffic to P2P traffic. This paper studies thoroughly on P2P systems and an interesting result was found about the scalability of P2P systems. The expected wide distribution of server load for P2P systems does not appear. The author also proposes that caching will improve the performance of those systems.

        The weakness of this paper could be on the experimental data. It would be better if a larger data collection could be made and data were not only from UW in and out traffic.

        Overall, this is a good paper with a very thorough analysis on Internet content delivery systems.

  • Next message: Tom Christiansen: "Saroiu et al, 2002"

    This archive was generated by hypermail 2.1.6 : Sun Nov 07 2004 - 16:03:33 PST