Abstract: In this paper, a scheme is presented, which builds high performance and high availability database cluster by using actively replicated TCP connections. The scheme forwards a technique that actively replicates a set of TCP connections by converting them into atomic multicasting and distributing read-only requests to different processing units. The technique is further used on MySQL database clusters and results in high performance and high availability. With the experiments conducted on the prototype systems, high performance on read-only database queries with little sacrifice on the update operations of the cluster is exhibited.