ANALYSIS OF TASK ASSIGNMENT POLICIES IN SCALABLE DISTRIBUTED WEB-SERVER SYSTEMS

A distributed multiserver Web site can provide the scalability necessa ry to keep up with growing client demand at popular sites. Load balanc ing of these distributed Web-server systems, consisting of multiple, h omogeneous Web servers for document retrieval and a Domain Name Server (DNS) for address resolution, opens interesting new problems. In this paper, we investigate the effects of using a more active DNS which, a s an atypical centralized scheduler, applies some scheduling strategy in routing the requests to the most suitable Web server. Unlike tradit ional parallel/distributed systems in which a centralized scheduler ha s full control of the system, the DNS controls only a very small fract ion of the requests reaching the multiserver Web site. This peculiarit y, especially in the presence of highly skewed load, makes ii very dif ficult to achieve acceptable load balancing and avoid overloading some Web servers. This paper adapts traditional scheduling algorithms to t he DNS, proposes new policies, and examines their impact under differe nt scenarios. Extensive simulation results show the advantage of strat egies that make scheduling decisions on the basis of the domain that o riginates the client requests and limited server state information (e. g., whether a server is overloaded or not). An initially unexpected re sult is that using detailed server information, especially based on hi story, does not seem useful in predicting the future load and can ofte n lead to degraded performance.