Server Replication in Interactive, Push-Based Data Delivery Networks

In a push-based system, updates that are made to objects are sent to clients, without them explicitly requesting these updates. The focus of this research is to develop a server replication protocol for such interactive, push-based networks. When a server gets overwhelmed and is serving more clients than it can handle, a replica is created. After the replication, a portion of the clients connected to the overwhelmed server are redirected to the new replica. The decisions (a) “at which particular server the replica is created” and (b) “which clients are redirected” can greatly influence the performance observed by the clients, the level of the service provided by the server, and the overall level of satisfaction with the functioning of the servers. In this paper, we identify a number of parameters that influence the selection of a new server and we develop an efficient and effective protocol, based on these parameters. The performance of the protocol is investigated through simulation. The protocol performs close to an ideal placement of objects and clients. On average a savings of 25% in was observed when compared with an uninformed strategy.