An Improvement on the Weighted Least-Connection Scheduling Algorithm for Load Balancing in Web Cluster Systems

Web cluster systems consist of a load balancer for distributing web requests and loads to several servers, and real servers for processing web requests. Previous load distribution scheduling algorithms of web cluster systems to distribute web requests to real servers are Round-Robin, Weighted Round-Robin, Least-Connection and Weighted Least-Connection(WLC) algorithm. The WLC scheduling algorithm, in which a throughput weight is assigned to real servers and the least connected real server is selected for processing web requests, is generally used for web cluster systems. When a new real server is added to a web cluster system with many simultaneous users, previous WLC scheduling algorithm assigns web requests to only the new real server, and makes load imbalance among real servers. In this paper, we propose a improved WLC scheduling algorithm which maintains load balance among real servers by avoiding web requests being assigned to only a new real server. When web requests are continuously assigned to only a new real server more than the maximum continuous allocation number(L), the proposed algorithm excepts the new real server from activated real server scheduling list and deactivates the new real server. And after L-1 allocation round times, the new real server is included into real server scheduling list by activating it. When a new real server is added to web cluster systems, the proposed algorithm maintains load balance among real servers by avoiding overloads of the new real server.