Virtual Machine Consolidation Strategy Based On Load Forecasting And Optimization

Nowadays,the demand of high-density commercial computing and scientific applications for cloud computing is increasing,which makes the construction of cloud infrastructure into a high-speed development.However,a large-scale cloud infrastructure consumes a lot of electric energy,and the cost of those power has been getting closer and closer to the cost of the infrastructure itself in recent years.In addition,energy consumption will produce a large amount of greenhouse gases such as carbon dioxide,which will aggravate global warming.Therefore,in order to reduce the energy consumption of the cloud computing center,virtualization technology is usually used to consolidate multiple services on one physical host as much as possible in the form of virtual machines to reduce the use of unnecessary hardware resources.In this paper,considering the different stages of virtual machine consolidation,combined with the historical data of the physical host,consolidate strategy is first proposed with the Kalman filter prediction model as the core.Then,the pheromone update rule of the heuristic ant colony system is used as the innovation point,focusing on the virtual Machine placement is optimized.Through the research on the existing prediction algorithm,we can find that the change of energy consumption has a certain correlation with time.Therefore,according to this characteristic,A virtual machine consolidation strategy based on Kalman filter is proposed.In the overload and low load detection phase of the host,we use the Kalman filter model to predict the host load energy consumption value at the next point in time,and combine with certain strategies to select the set of overload and low load hosts.In the virtual machine selection stage,we first obtain the CRO value which is virtual machine CPU resource overheating rate,and then select the virtual machine to be transplanted based on the combination of the three indicators of memory and network utilization.In the stage of virtual machine placement,we combine the best fit decreasing(BFD)algorithm with Kalman filter prediction,so as to reducing the judgment of virtual machine placement,avoiding frequent placement-migration operations of physical hosts,and improving placement efficiency.In the aspect of virtual machine consolidation,the main idea is to find a reasonable relationship between the virtual machine and physical hos,so that the energy consumption of the entire system is the lowest.Therefore,we use a heuristic algorithm-ant colony system to search for the global optimal solution.The strategy regards the virtual machine placement problem as a multi-objective optimization problem related to energy consumption and the number of migrated virtual machines.By simulating the foraging process of the ant colony to find the best placement mapping relationship between the virtual machine and physical host,the main improvement points are the global pheromone update rule,heuristic factor setting and application of Kalman filter model in prediction.so that the ant colony can search in a wider range and avoid excessive converge early.After a limited number of iterative searches and information exchanges,the ant colony obtained a global better solution to the virtual machine placement problem.This paper uses Cloud Sim platform to simulate the above two strategies,and measures the performance of the load predict algorithm and other comparative algorithms in improving system performance according to the performance evaluation index.