A replica management strategy based on MOEA/D

As distributed storage systems have been used more and more widely nowadays, data replication management has become a hot research topic. Although replica technology can increase data availability and improve load balancing, an unreasonable replica management strategy will waste too much storage and increase the energy consuming sharply. In this paper, we developed a replica management strategy based on MOEA/D, in which we take five factors affecting replication decisions into consideration such as system data unavailability, network traffic, disk performance, load balancing and energy consuming. By using the improved MOEA/D algorithm, we get a Pareto solution sets, then an elite selection strategy based on individual density was proposed to pick out the suitable solution from the set. At last, we adjust the replica dynamically depending on the file access record log. Those steps are gathered to form the final strategy: A Replica Management Strategy Based On MOEA/D(MDRMS). Series of experiments are designed to demonstrate the validity and universality of the strategy. The experimental results show that MDRMS performs better than HDFS(Hadoop Distributed File System) and DRPMO (Dynamic Replica Placement based on Multi-objective Optimization), especially in load balancing and energy consuming.