Special issue on Internet architecture technology papers: A nonprobabilistic server selection method based on periodically obtained load information for Web server clusters

For a system comprised of multiple Web servers that provide the same content, this paper proposes a method for the dynamic selection of servers that process requests, based on periodically obtained load information on servers. Previously proposed methods select servers probabilistically based on the weights calculated from server load information, and consequently, a server with a heavy load is selected by chance, often resulting in decreased performance. In contrast, the present method, instead of employing probabilistic server selection, selects servers by a round-robin scheme after eliminating any difference in load among the servers. A comparison of performance based on simulation studies with the conventional method indicates that the proposed method is immune to the impact of the intervals at which load information is acquired or the number of servers involved, and that it produces a low average response time even under heavy load conditions. © 2006 Wiley Periodicals, Inc. Electron Comm Jpn Pt 1, 89(10): 1–12, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecja.20309