Design and Implementation of a Scalable Web Server Cluster

Abstract

In this paper, we describe the design and implementation of a Web server cluster. We first discuss the design criteria of a good server cluster. Based on the criteria, we present the design of our proposed Web server cluster which consists of a central server and multiple Web servers. The central server stores the whole document set, while the Web servers only cache a subset of the documents. The cache mechanism is based on our proposed probability- based replacement (PBR) algorithm which optimizes both the storage requirement and network I/O. When a Web server received a client request for a document cannot be found in its local cache, it asks the central server for that document. The main advantages of this design are that it can be scaled with least effort and integrated to existing systems easily. We then present the implementation details of this cluster that has been built and deployed in a real case. We also compare the maximum throughput between the system using the proposed cluster and the system using mirrored Web servers. Both analytical and simulation results show that, with 10% document caching, the maximum throughput of our cluster is nearly the same as that of mirrored Web servers.