URI-sniffing based load balancing on clustered web servers

Abstract

With the rapid growth of World Wide Web, popular sites experience continuous and sometimes disproportionate increase in client requests.Growing demands on scalable systems have led many sites to deploy cluster-based server architecture. As the Web is popularly used as an interface to new applications such as e-commerce, the proportion of resource-intensive dynamic content is also increasing. Processing a dynamic content request on the server usually requires one or two orders of magnitude more system resources compared to processing a static content request. High variation of content increases the variation of resource requirements, which may cause skewed cluster utilization. Bursty requests and high variation of content has placed heavy demands on scalable routing mechanisms and sophisticated load balancing policies which can provide scalable performance and desired level of service quality. Recently, layer-4 (L4) Web switches have been widely used as load balancing front-end routers for Web server clusters. Due to the low-overhead TCP routing mechanism, they can scale to large clusters. However, there is no means for balancing load among the clustered back-end servers besides distributing new connections across them. Therefore, timely and accurate estimation of each server`s load is vital to keep the servers evenly loaded. For this purpose, the typical L4 switch requires some form of load metrics feedback from the servers. Unfortunately, the insufficiency, measurement overhead, and feedback delay of load metrics often cause misestimate of server load. This may often result in significant dynamic load imbalance among the servers when the variation of requested content is high, which increases user-perceived service latency and degrades server performance. As an alternative, layer-7 (L7) Web switches intend to provide high quality of service by dispatching client requests to suitable servers on the basis of the requested content, denoted by Uniform Resource I...