Research On Virtual Machine Migration Mechanism Based On Lord Forecast In Cloud Computing

The current cloud computing industry is developing rapidly with more and more users choosing to move their operation system to the cloud computing system, which could save a large sum on basic installment expenditures as well as avoid the onerous system design and maintenance works so as to focus more on their professional fields. Owing to its business service mode of dynamic expansion and serving according to needs, the cloud computing system well fits in the demand of pursuing low cost information services and low resource consumption for the current society and thus draws high attention from all sides.The technology of virtual machine migration as the key technology at the bottom of resource management in cloud computing system, its performance plays a vital role in the overall performance of the cloud computing system. The current major strategy of virtual machine migration is to respond passively, namely the cloud computing system responding to user’s requests in actual time and the virtual machine migration conducted according to the set strategy to achieve the continuity of the service. With the arrival of cloud computing application, many task requests feature a periodical change: when it reaches the peak value, the amount of the requests increases dramatically exceeding the cloud computing system’s momentary responding capacity, thus resulting in obstruction, which will affect users’ experience. Targeting at this issue, the current paper designed a virtual machine migration mechanism based on “ active prediction”, forecasting the changing trend of cloud requests amount in advance to ameliorate the drawbacks of the current passive response to migration tasks. This mechanism through migrating the virtual machine in advance and adjusting migration strategy realizes the goal of improving the migration performance of the virtual machine and breaking through the bottleneck of the current migration strategy’s performance. The following are the main works:According to the tendency, seasonal and randomness characteristics of the periodical task requests as well as the affection on the system’s performance when the requests reach the peak value, we designed a LF-HW server hotspot testing model for the periodical tasks. The model can predict the peak amount value of requests in the next period of time through analyzing the pattern of task requests history. The virtual machine would be migrated in advance based on the predicted value and actively respond to the coming peak of requests. LF-HW model can estimate the amount of task requests under the cloud computing condition via the Holt-Winters cubic exponential smoothing forecasting model in the prediction module and respond to the predicted amount of task requests with the designed prediction algorithm, determining whether it is necessary to conduct the pre migration of the virtual machine and calculating the informationsuch as their quantity, location and timing etc. We conducted the simulation experiment for the LF-HW model by expanding and compiling the cloud computing simulation platform CloudSim, to verify the effectiveness of the model and the accurateness of the prediction algorithm.For the problems of poor performance and combinatorial optimization cannot be achieved to cut down the expenditure existing in the strategy of randomly selecting the point where few idle resource were left for migration in the virtual machine migration strategy, we designed the virtual machine migration selection strategy which is based on the improved particle swarm optimization algorithm. This selection strategy avoids the resources occupied by the virtual machine exceeding the maximum limitation of the server to reach the goal of combinatorial optimization by the method of defining the fitness of the virtual machine that is matching the distance value and the server’s remaining property, as well as utilizing the server’s remaining property avoidance sheet. Through the simulation experiment, it also turns out that the improved particle swarm optimization algorithm possesses fairly well optimization ability in choosing batch points of migration.Through setting up the actual desktop cloud management system, we established the server hotspot testing model for periodical task requests and the selection strategy based on the improved particle swarm optimization algorithm. By deploying different service at the points, it proves that the method proposed in this paper excels the original system in total migration time,response time and the amount of data transmitted. It also turns out that the proposed virtual machine migration mechanism can effectively enhance the system ’ s performance in virtual machine migration and bears great load balancing ability under the cloud computing environment.