The number of users and hosts connected to Internet has been growing at an exponential rate, resulting in increases in network traffic, server load and data retrieval latency. The basic Internet client-server model wastes network bandwidth and stresses the popular servers by sending the same data repeatedly over WAN links so that it increases the response time for Internet users. One way to solve these problems is resource replication. However, when a resource is replicated, clients then need additional capability to find a "good" server of that resource.
In this thesis, we suggest a mechanism for finding nearby replicated servers without a priori knowledge of server location or network topology. By grouping replicated servers, a client accesses replicated resources as a whole. In addition, consistency among original resource and replicated ones is maintained efficiently by exploiting group communication mechanism. We developed the group management system which provides server group creation/deletion, membership control and consistency control among group members. A prototype system has been implemented using IP Multicast and tested on the MBone.
