Dynamic scheduling of Web server cluster

Abstract

Web servers have recently been composed of a server cluster for keeping good quality of service. LVS (Linux Virtual Server) is one method for building a server cluster in Linux and has its own load balancing mechanism. The existing load balancing mechanism, however, often reaches a limit and fails to keep the load in balance, particularly in electronic commerce sites that use typically dynamic script files more than static HTML pages. In this paper, we suggest a new mechanism that overcomes the limit and discuss how to define the load of each real server node in the server cluster and determine the limit of the load. In the suggested mechanism, the weight table of a load balancer is updated according to variation of the load and distribution of service requests is controlled according to the table. The new mechanism shows better results to distribute the load evenly among the real servers than the existing mechanism, when many scripts are included.