Virtual Machine Placement Optimization Technology Based On Forced Insertion Heuristic Algorithm

With the widespread spread of the Internet and the increasing of network flow scale,while the hardware resources required for network applications continue to rise,the cost of traditional enterprise self-configuring servers has also increased.Cloud platforms become the better choice of small and medium size enterprises.However,High energy consumption brought by the cloud platform has become an urgent environmental problem.Virtual machine technology allows a single server to run multiple operating systems at the same time,giving the ability that supply hardware resources and independent environments to multiple client applications at the same time to any server.Virtual machine placement algorithms calculate the deployment solution that using least servers according to the resource owned by servers and the resource requirements of client applications,thereby reducing resource consumption.At present,deployment solutions obtained by various virtual machine placement algorithms can effectively reduce active server numbers.However,running time of the existing algorithm is hard to meet the needs of the production environment.Aiming at that,thesis proposes a heuristic algorithm based on the forced insertion operation.Main contents are as follows:（1）By analyzing the optimization results of existing virtual machine placement algorithms,found that compared with low load servers,the average resource requirements of virtual machines deployed by high load servers are smaller.Because of plenty of existing algorithms focus on the server payload rate,if update operation reduces the payload rate of the high load server,existing algorithms will generally cancel the update,resulting in the solidification of the high load server.In addition,by setting up marked high load server in each individual before the initiation of the existing algorithm,it is observed that during the iterative process if server load rate is high enough,its virtual machine deployment structure is difficult to change,resulting in the difficulty of further optimization in the later stage of the algorithm.（2）To solve the above problems,thesis proposes the forced insertion operation.Maintain or further increase payload of high payload servers without affecting the payload of the high load server as much as possible.During the exchange step in the forced insertion operation,multiple virtual machines in the high payload server are exchanged with the large resource requirements virtual machines deployed in the low payload server,changing virtual machine deployment structure by the high payload server while the impact on the server payload is as small as possible.Filling step in the forced insertion operation places multiple virtual machines obtained from the exchange operation into other high payload servers,maintains or increases the load of the high payload servers,and further reduces the payload of the low payload servers.（3）Thesis proposes the adjustment operation.Adjustment operation uses reconstruction strategy to redeploy all virtual machines of the placement solution and reduce the number of servers.Experimental data show that adjustment operation using reconstruction strategy based on random strategy can quickly obtain a reasonable placement solution with less servers.（4）The proposed forced insertion heuristic algorithm maintains the payload of high payload servers on the one hand,and reduces the payload of low load servers on the other hand,accomplish the forced extraction of large resources requirement virtual machines in low payload services and exchange it with the virtual machine in high payload servers.After the forced insertion operation,payload of the low payload servers are further decreases.Algorithm uses the adjustment operation to select the low load server in the placement solution,so as to quickly obtain a reasonable placement solution with less active servers.（5）Experimental verification and result analysis of homogeneous and heterogeneous server scenes.Use the proposed algorithm and the contrast algorithm to conduct multiple scene experiments,verify that the proposed algorithm has the ability to perform efficient and fast in homogeneous and heterogeneous server scenarios.And further conduct the energy consumption experiment and parameter debugging on the proposed algorithm.Thesis compares the proposed algorithm with RGGA and OMEACS through experiments.The experimental results show that the proposed algorithm using the strong interpolation principle obtains a close-to-ideal solution in a short time through a large disturbance strategy in the optimization process,and shows good stability.