A scalable and adaptive Web client-server architecture

The World Wide Web has become the most popular way of exchanging various kinds of data and performing applications such as electronic commerce. Unfortunately, the increasing demand for Web service cannot be satisfied by the increase in the network and computer capacities. Therefore, there exists an immediate need for software solutions to improve the Web service. This thesis describes a new Web client-server architecture that aims at minimizing the average response time observed by the Web users. The proposed architecture applies an adaptive replication algorithm that brings the Web resources as close as possible to the users who access those resources frequently. It also provides mechanisms for load balancing and increasing the reliability of Web service. The servers are responsible for replication of Web resources, and informing the clients about the locations of replicas. The clients are capable of selecting the best replica location to satisfy a particular request.; We have implemented a prototype of the proposed client-server architecture, called Web++. Web++ clients and servers can run on top of the existing Web servers and browsers, and enhance their capabilities without requiring any modifications to the existing Web infrastructure. Web++ clients can be downloaded and executed on Web browsers transparently.; The smart clients of Web++ are capable of performing other tasks in addition to selecting the best server from which they could request a particular resource. One such task is batch transmission of Web resources which means submitting a requested resource together with all of its embedded resources in one response. Batch resources transmission reduces the response time by minimizing the number of request and response rounds between a client and a server.; It is possible to improve the Web service performance further by using clusters (groups) of relevant resources as the unit of replication. We propose a distributed clustering algorithm that identifies relevant resources from user request patterns. The communication overhead can be reduced by using clusters as the replication unit since a client can request multiple resources from the same server through the same connection.