Protocol for pseudo-active replication in wide-area networks

Abstract

We discuss how to replicate a server in order to obtain fault tolerant services. In the active replication, all the requests from a client are performed by all the server replicas in the same order. The replicas are rather placed on heterogeneous computers interconnected in a wide area network: (WAN), e.g. the Internet, where each communication channel has different message transmission delay and bandwidth. Hence, the response times of the replicas observed by the clients are not the same. We newly propose a pseudo active replication where the client does not wait for other responses after receiving one response and slower replicas do not perform every request performed by the faster replicas.