Presenting a prototype for pull based load balancing of web servers

Abstract

The traffic and the number of users on the Internet are increasing, and service providers must respond to this demand if they are not to loose both customers and revenue. In order to provide a satisfactory level of QoS, the providers must increase server performance by aggregating multiple physical host into a web farm, which by working together, sharing the total load of the requests, will act as one unified server. Various algorithms and technology exists for performing this load sharing, and this thesis will introduce a new approach in this field. In the traditional load balancing (LB) strategies, the requests are pushed to the individual servers, which are passively forced to accept the requests. The new approach presented in this paper, differs from the traditional techniques, in that the servers themselves are more involved in the decision making associated with load balancing. Requests are stored in a central queue, and the individual requests is processed by the servers at the server’s own convenience, leaving the server with full control over it’s own resources. The paperwill present and compare this algorithmwith the traditional load balancing algorithms, and investigate possible benefits of this strategy.