WWW Proxy Traffic Characterization with Application to Caching

This paper identifies invariants that hold across a collection of ten traces representing traffic seen by caching-proxy servers, finding a strong connection between access rate from hour to hour and sources of Web self-similarity.

Abstract

Characterizing World Wide Web proxy traffic helps identify parameters that affect caching, capacity planning and simulation studies. In this paper we identify invariants that hold across a collection of ten traces representing traffic seen by caching-proxy servers. The traces were collected from governmental, industry, university, high school, and an online service provider environment, with request rates that range from a few accesses to millions of accesses per hour. We also show that the examined traffic is semi-similar. We explore sources of Web self-similarity and we conclude that a strong source is the periodicity in the users behavior. The tests revealed that there is a strong connection between access rate from hour to hour. We also report the hit rate and weighted hit rate obtained by running a trace driven simulation on the workloads to simulate a proxy with infinite cache, similarly, accesses to unique servers and URLs are a small portion of the total. By considering these characteristics of traffic we can improve the utility of caching for WWW clients.