Study Of Virtual Machine Placement Strategy Based On Dynamic Load Characteristics

To ensure the user’s service level agreement (SLA) and improve the efficiency ofresource utilization of data centers is the very focus in the research of cloud computing.Virtualization technology is the key technology of resource management in cloudcomputing, and the virtual machine placement strategy is an important means ofvirtualization resources management. it includes three steps, which include thedetermination of migration time, the choice of the virtual machine and the mappingfrom virtual machine to the host node. The primary goal of this paper is to design anintegral and efficient virtual machine dynamic placement strategy which contains thethree steps above. The strategy will be more suitable for the cloud computingenvironment.In order to determine the migration time, this paper proposes a mechanism basedon the prediction of load history, by predicting the future load information onreasonable time to reduce the number of migration greatly. For selection problem ofvirtual machine migration, This paper proposes a "filtration+resource matching"multi-step selection mechanism. Firstly, the mechanism filters out all the virtualmachines which is not suitable for migration. Then, make pairing matches between virtualmachines on the basis of the principle of time of resources complementary. By reducingresource competition between the virtual machines to reach the purpose of improvingthe efficiency of resource utilization, and greatly reduce the virtual machine migrationoverhead caused by migration. In view of the virtual machine to the host node weightmapping problem, this paper proposes a mapping algorithm which is based on adaptivemutation particle swarm optimization algorithm (AMPSO) and load characteristics thatconsiders the node resource heterogeneity. The algorithm aims to maximize theresource utilization, and minimize the SLA violation rate at the same time ensuringload balance of the data center. The result of the algorithm is verified by simulationexperiments. Through the experiment and comparison for the three steps of thisstrategy, the result indicates that the methods of the paper can solve the problem ofdata center resources redistribution management more efficiently.