From: Chuck Reeves (creeves_at_windows.microsoft.com)
Date: Mon Mar 08 2004 - 15:08:51 PST
The paper, "Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload" was written by a number of researchers at the University of Washington in 2003. It presents an analysis and conclusions of a study done to analyze Kaza peer-to-peer file sharing traffic at the University of Washington over a number of months.
The researchers found that the Zipf models typically used to characterize content usage for web services did not accurately predict the behavior of Kaza users. The study presents a new usage model dubbed "fetch-at-most-once" as a new model appropriate for situations where content is considered immutable and new content is made available through the introduction of new documents. This is unlike web content in that new content on the web is introduced through updates to existing documents. This difference results in a lower level of interest in documents found at the popular end of the curve than in Zipf. The authors also prsent some analysis on the locality of Kaza relative to the University network. Simulated analysis on the introduction of caching into this environment indicated that a dramatic reduction in traffic on University-to-web boundary was possible.
Studies such as this help me better understand how web service systems should be built. The new model presented here indicates that a change in our thinking may be necessary to accomodate the way the web will be utilized in the future.
This archive was generated by hypermail 2.1.6 : Mon Mar 08 2004 - 15:09:19 PST