Jerrymouse: a tool for a flexible and dynamic distribution of web service requests

This paper presents a novel architecture for distributing web service requests on clusters of servers. The architecture facilitates a transparent dynamic distribution of requests according to a range of specified policies. This enables a flexible performance in respect of different objectives, services and platforms (typically based on server workload). The architecture has been successfully demonstrated with a prototype implementation (called “Jerrymouse”). Our preliminary results with Jerrymouse indicate stable behaviour and worthwhile performance gains (compared with Apache HTTP Server). A specific policy to deliver reduced cluster electricity savings has also been successfully implemented