Cluster architecture with lightweighted redundant TCP stacks

Availability of the service provided by the cluster system is greatly emphasized in today's system domain. We proposed a new technique, named redundant TCP stacks (RTS), to enhance the connection-level reliability and availability of the services provided by cluster system. In this paper, we discuss a faster synchronization algorithm, named send when the fastest response (SWFR) to alleviate the overhead introduced by RTS. We evaluate the new algorithm by comparing it with the original algorithm of send when the minimum updated (SWMU) and the traditional TCP communication. By experiments, we find that RTS with SWFR algorithm minimize the sacrifice on network performance. The performance of a cluster with two servers working in RTS scheme is satisfactory enough for practical usage while delivering high availability.