AR-TCP: Actively Replicated TCP Connections for Cluster of Workstations

In this paper, we propose a novel scheme, called actively replicated TCP (AR-TCP), which transparently improves the service and data availability of cluster systems at TCP connection level. In this scheme, TCP connections are replicated and synchronized among all workstations of a cluster and can failover to healthy parts during failures. Moreover, request messages resulting in data exchange are delivered by the workstations atomically to guarantee data consistency. Read-only request messages are extracted and executed on only one of the replicas to improve the performance of simultaneous access. As an application, we build a fully replicated MySQL cluster based on our proposed scheme. The experimental results of the prototype implementation show that the cluster has small performance penalty on communication, high simultaneous read-only query performance and a small performance penalty on update operations