Scaling server selection using a multi-broker architecture

Server replication is a common approach to improving the scalability of a service on the Internet. For this approach, the task of finding an appropriate server from a set of replicas is a critical issue. We have proposed in a previous work an architecture where a broker is used to provide server selection on a per session basis. When the number of servers and/or the number of clients becomes large, a single broker may not have sufficient capacity to handle the load. An extended architecture based on the replication of brokers is therefore considered. We first discuss alternative organizations that support access to multiple brokers and the needed cooperation between brokers in order to achieve server selection effectively. We then propose a server selection policy for our multiple broker architecture and evaluate its performance by simulation.