Research On Strategy Of Virtual Machine Resource Allocation In Cloud Data Center

As a future trend of calculation model and a core of new generation ofinformation technology and commercial model change, Cloud computing is attractingmore and more attention of the industry and the academic, and has broad marketprospects. At present, almost all IT giants have studied the cloud computing fromdifferent directions according to their areas of technology and marketing strategies.With the development of cloud computing, the cloud data centers are becomingincreasingly large scale. Data center form a huge virtual resource pool byvirtualization technology, but due to the lack of effective data center resourcemanagement mechanism, the virtual machine can’t have a reasonable allocation. Forexample, with the operation of the system and the change of customer service load,virtual machine placement becomes disorderly, which is not a good response to peakload problem caused by user burst access, leading to the user’s experienceperformance degradation. Also, if we are able to carry out a unified plan to deployvirtual machines from the global scope, regularly, we can improve the utilization ofdata center resources to some extent. Therefore, it is has important academic andpractical significance to research data center resource management, especially in thevirtual machine resource management.After in-depth discussions and careful study of the existing algorithms, this paperpresents dynamic virtual machine management algorithm based on load predictionand virtual machine placement algorithm based on improved genetic algorithm, andgains several innovations and achievement. The major contributions of thisdissertation are as follows:1. Dynamic virtual machine management algorithm based on load prediction:dynamic consolidation of virtual machines in cloud data centers involves threestages-when a virtual machine migration needs to happen, selection of virtualmachines that should be migrated from a host, finding a new placement of the virtualmachines selected for migration from the overloaded and underloaded hosts. Thispaper presents a dynamic virtual machine management algorithm based on loadprediction, which use exponential smoothing model to predict future time loadconditions, combined with the maximum correlation strategy and power aware first fit decreasing, achieving dynamic balance the load of entire data center. The simulationresults show that the algorithm can reduce the energy consumption of data centers andservice level agreement violations, effectively increasing the overall resourceutilization of data center as the core of the cloud infrastructure.2. Virtual machine placement algorithm based on improved genetic algorithm:Most studies use the heuristic algorithm to solve the problem of virtual machineplacement, but the traditional algorithms can only give local optimal solution of theproblem, lacking global optimization ability. This paper presents a virtual machineplacement algorithm based on improved genetic algorithm. The innovation is that inaddition to considering the energy consumption of physical hosts, but also taking intoaccount the energy consumption of the data communication between virtual machines.In addition, it has the infeasible solution repairing mechanisms and globalconsolidation mechanisms. Experimental results show that compared to heuristicmethod and genetic algorithm, our algorithm can reduce the number of physical hostsand energy consumption.