Replica Placement in Cloud Storage based on Minimal Blocking Probability

Data replica placement is an important issue in the cloud storage system, which includes two key points: how to determine the replica factor and how to select optimal Datanode to store replica. With the provision of file store service and file access service for large scale users, the Datanode in cloud storage system is regarded as services site. According to the basic principle of queuing theory, the replica placement algorithm based on minimal blocking probability (BPRA) is proposed from the view of resource competition as a dynamic replica placement algorithm. The BPRA algorithm calculates the minimal number of replica according to the file available requirement and dynamically adjusting replica factor according to the file access frequency. In addition, the blocking probability is computed by each Datanode respectively and reported to the control node (namely Namenode). The simulation results conclusively demonstrate that the replica placement algorithm based on minimal blocking probability can effectively achieve the load balance of cloud storage system, reduce access skew and resource competition, decrease 20%~60% file access latency versus Hadoop Distributed File System (HDFS) default replica placement algorithm.