Research On Virtual Machine Consolidation Strategy Based On Load Prediction

With the development of globalization,the cloud computing industry,as an important part of the Internet economy,has shown a vigorous development trend,but at the same time,it is also facing many challenges,especially the high energy consumption of cloud data centers.The virtual machine integration strategy is one of the main ways to solve the energy consumption problem.Its core is to use the virtual machine migration technology to move the virtual machine on the low-load host,and then sleep or shut it down,thereby reducing the active physical host.Reduce energy consumption in cloud data centers.However,aggressive integration strategies may result in large-scale virtual machine migrations,or some hosts are overloaded due to sudden load increases,which not only leads to a decline in service quality,but also may generate more energy consumption.This article uses a predictive model to serve the virtual machine integration process,and proposes a virtual machine integration strategy that adapts to cloud data centers.The main tasks are:In the host overload detection stage,an overload host detection(GM-OD)algorithm based on the gray GM(1,1)model and an ES-OD algorithm based on the exponential smoothing model are proposed.The process is to predict the load value of the host in the next period according to the gray GM(1,1)model and the exponential smoothing model,and compare the predicted value with a preset threshold.If the predicted value exceeds the upper threshold,the host is judged to be overloaded.Repeat this process to determine the set of overloaded hosts.In order to solve the high-load work pressure of overloaded hosts,a virtual machine selection(RUA-VMS)algorithm based on resource utilization awareness is proposed in the virtual machine selection phase.The core idea is to select the virtual machine on the overloaded host that occupies the maximum CPU resource utilization of the host,Add it to the set of VMs to be migrated,and repeat this process until the host is no longer overloaded.In the virtual machine placement phase,an optimal adaptive descending order(RRMin-BFD)algorithm based on remaining available resources to minimize constraints is proposed.Sort all the virtual machines to be migrated in descending order according to the amount of CPU resource requests.According to the idea of "use everything",the host with thesmallest available resources after receiving the virtual machine is taken as the best target host.Repeat this algorithm and wait for all Migrate virtual machines for reasonable placement.According to the three stages of overload host detection,virtual machine selection and placement integration,we get a virtual machine integration strategy based on gray GM(1,1)prediction and a virtual machine integration strategy based on exponential smoothing prediction.This paper uses the Cloud Sim platform to perform these two algorithms.Simulation experiments and comparisons in terms of ensuring service quality and reducing energy consumption based on performance evaluation indicators and several other algorithms.