In this paper, we propose a novel scheme that can be used to build high performance and high available database cluster. This scheme employs a middleware approach, which can replicate the databases on the server nodes of a cluster and keep them consistent during the fly. Read-one-write-all strategy is used to execute the read only queries of the concurrent transactions in parallel so as to improve performance. Compared with other replication schemes, such as PostGreSQL-R, this scheme does not need to consider the underlying details of database management system, and thus has the advantages of flexibility and portability. The simulation results show that this method yields better performance than traditional database replication techniques and lower abort rate.