Abstract: In order to guarantee the stability of the system performance and the high Qos(quality of service) of users, a new method based on the HDFS (Hadoop Distributed File System) was proposed which including a job type classification method and a dynamic replica manage mechanism. The method uses the job type classification method to select the I/O intensive job, in order to achieve more accuracy of the classification taken the heterogeneity of the jobs into consideration. For the classified jobs, a dynamic replica manage mechanism was used to determine whether to increase or decrease the number of copies on the specific data node. For a test of a cluster with 1 namenode and 20 data nodes, the method has a high performance. The theoretical and experimental analyses in this paper prove that the proposed method has the ability to improve the performance of HDFS effectively.