Maintaining consistency among replicas is an important technique for improving the efficiency of data grids. In order to improve replica consistency management, one fast reading and writing method is proposed, where two update algorithms is applied to accelerate Maintaining consistency among replicas. The first layer is synchronization writing form one peer to many peers at the same time, thus avoiding repeat read and writing of replica from the primary node. In second algorithm the node with good bandwidth is elected as new master node to enlarge the number of concurrent update streams. The experimental results show that the proposed model can reduce complete time compared with lazy copy, positive copy in some scenes.