A dynamic load scheduling algorithm for Web servers

As used increasingly the embedded objects and the database searching tasks in Web pages, there is larger difference among the loads of different server in a cluster system, it becomes more difficult for a heterogeneous Web server cluster to achieve high performance. In this paper, we present a dynamic algorithm on load scheduling, namely multiplicative decrease in critical area. For each of the servers in the cluster, the algorithm can more accurately evaluate the current load state by using the equivalent load-alternant and can more efficiently restrain the occurring of the reject service phenomenon by using a special MDC operator. Besides, we apply a method of random distributing base-probability to assign each request to an appropriate server in terms of their weight. All the parameters that will be used in our algorithm could be acquired by test. We also provide improved approximation results of our algorithm for the case where documents are consisted of relatively many embedded objects or database searches and lots of requests are arrived at the dispatcher synchronously.