Analysis of design alternatives for reverse proxy cache providers

Abstract

Reverse proxy caches are used to provide scalability and improved latency to popular sites on the Web. In this paper we provide analytical performance models for distributed reverse proxy cache architectures, and study the trade-offs between various design alternatives. Specifically, we consider static and dynamic assignment of proxy cache nodes to Web sites, with different levels of sharing of proxy caches among Web sites. Innovative modeling contributions have been introduced to handle real design constraints, such as bounded cache size and bounded processing power, and different characteristics related to the hosted objects, including reference rates, popularity distributions and update rates. In the analysis we have modeled both system steady state as well as transient interaction between proxy sites and Web sites. We have found different trade-offs between various design alternatives depending on characteristics of the Web site workloads.