The Research Of Live Data Migration Based On Elastic Load

Cloud computing let us have a better utilization of distributed resources, and economize the cost as well, but on a technical side, the hottest issue in the cloud environment is to realize the elastic load.Unpredictable data accessmay lead to the dynamic change ofload in the cluster, in this case, we realize elastic load balancing and dynamic management of service-oriented computing resourcesby live data migration. Live migration migrate data dynamically among the nodes with low overhead, it is developed to meet the characteristic of elastic load balancing in cloud database, and it can realize a lightweight elastic load. Now there are two main methods to realize live migration: iterative migration and copy-on-write migration, and there is less attention paid to the selection of the data and the destination node, and the control of the migration process. We hold that data migration include migration decision and migration execution, a good migration decision can save the migration time and reduce the migration overhead, but there are few researches about migration decision.In this paper, we study particle swarm optimization and data migration technology on the basic of the above research background, and glue them together to solve the problem of migration decision. Since the parallel implementation of traditional PSO is very time consuming in our scene, we realize parallel PSO through stream computing technology to save time and to improve the algorithm performance, and we design and accomplish Stream-based Particle Swarm Optimization(SPSO). We apply SPSO in migration decision that can meet the requirements of real-time, avoid disk reads and writes, and avoid time consuming. We propose a data migration method based on SPSO migration decision, which relies on our migration architecture. In this method, we can realize the real-time monitoring of node load, analyze the type of load, enable migration decision to generate the best migration plan, distribute the requests during the migration to ensure the accuracy of the data, and control the speed of the migration to ensure the performance of the source node and destination node, respectively. The experimental results of our method are given: 1) SPSO algorithm has better performance of accuracy and stability than PSO algorithm in migration decision, and it has shorter running time, the migration overhead of migration plan that SPSO gives is smaller than PSO algorithm as well. 2) During the data migration, the failure of write data operation significantly less than other methods, and the response time is obviously decreased after the start of the migration. 3) We analyze the load state of source node and destination node through the whole process of migration, we found that our method has relatively small impact on the performance of the source node and destination node, the operation of migration balances the load well between source node and destination node.