Abstract: HDFS takes random storage strategy,if cluster has some cheap nodes,it is possible to make high frequency data store in the low processing performance nodes,causing a long time access and poor efficiency.To solve these problems,an improved scheduling strategy of hierarchical storage about replication was proposed.In order to reduce the number of replication scheduling,firstly,the information of data node from CPU load,memory load,network load,sto-rage load and network distance are used to evaluate node availability.Secondly,the optimal one is selected.Accessing frequency and hardware configuration are used to realize the replication scheduling.The response rate of cluster is improved by making high frequency data store on the high processing performance and high configuration node.The experimental results show that the strategy can improve access efficiency of replicas and local balancing for data storage in the heterogeneous clusters.

Key words: Cloud storage,HDFS,Hierarchical storage,Replication scheduling