In this article the authors propose site-based mapping schemes for parallel proxy servers that forwards all requests targeting the same Web site to the same proxy server. The use of parallel proxy servers requires a mapping scheme to map each client request to a corresponding proxy server. The basic idea of site-based mapping is that all Web pages belonging to the same site will be mapped to the same proxy server. The authors propose a site-based mapping scheme to take advantage of persistent connections. This scheme can reduce the number of connection establishments, which in turn saves network bandwidth and shortens user-perceived delay. To gain the advantage of the persistent connection feature, subsequent requests must be sent to a site shortly after a connection to it is made. More requests using the active connection will make the connection better utilized. Unfortunately, the commonly used mapping schemes are all URL-based. In these schemes, the requests targeting the same site will probably be mapped and forwarded to different proxies, resulting in multiple connection establishments to the same site. This certainly would not optimize the use of persistent connection. With the site-based scheme, each proxy is configured to be responsible for a particular set of Web sites. All the requests targeting this set of preconfigured Web sites are forwarded to the same proxy.

The article discusses policy selection for Web caching which is a significant part of the present infrastructure of the World Wide Web. Web caching is significant as it reduces the number of requests reaching Web servers and overall network activity. A cache server stores Web objects locally for the use of future requests to those objects and due to the finite size of cache, a cache replacement policy is needed to manage cache content. This policy will determine which object is to be removed from the cache to make room for the incoming object. The replacement policy has the goal to make the best use of available resources, such as disk space, processing power, server load, and network bandwidth. There are some policies that favor one resource at the expense of another. Performance metrics are used for the evaluation of a policy and a policy may perform very well in terms of one metric while poorly in terms of another. The common metrics, which are used for policy evaluation, include, hit rate, byte hit rate, CPU utilization, and latency reduction.