Research On Load Forecasting And Balancing In Cloud Environment

With the rapid development of information technology,the processing model of the computer has undergone tremendous changes.From the mode that all tasks are delivered to a large processor,to the distributed task processing mode based the network,and it becomes cloud computing mode at present.In the initial stage which is a single processor mode,due to the restrictions of its processing capacity,the tasks always have a long waiting time,and then the efficiency also is low.With the development of network technology,the servers start to provide the services by the cluster.However,when the cluster is configured to deal with high load situation,it will have the resources idle and waste if the load of the cluster is low,resulting in increased maintenance costs of cloud agents.The cloud computing service mode means the service resource virtualization,users do not pay attention to service's implement.The management,scheduling and maintenance of the resources and the other works are in charge by the special personnel,so the cloud computing is essentially to provide users a kind of on-demand processing service which is similar to the traditional water and power service.In addition,the cloud computing is designed for users to maximize the use of virtual machine resources at any time and place through the network.In recent years,with the promotion of academia and enterprises,the research and application of the cloud computing problems are showing a trend of rapid development.Due to the startup time of the host,resource allocation time,task scheduling time and other factors,there is a delay problem in the service which is provided to the user in the cloud environment.Therefore,workload prediction is an important way of energy optimization in cloud environment.In addition,due to the great fluctuation of cloud workload,the prediction difficulty of the model is increased.This paper presents a prediction model(Hybrid Auto Regressive Moving Average model and Elman neural network,HARMA-E)based on autoregressive modal and Elman neural network.Firstly,using ARMA model to predict,and then using ENN model predict to ARMA model prediction's error for modifying prediction error of ARMA.The experimental results show that the proposed method can effectively improve the prediction accuracy of the host workload.Implementing load balancing is also an important aspect of improving resource utilization and quality of service(QoS)of virtual machines.At present,most of the load balancing algorithm make virtual machine CPU,memory and other resources rate as the optimized goal,without considering the execution time of the total tasks are unbalanced on the virtual machine led to increases the makespan of the total tasks.This paper proposes the PSORF(Particle Swarm Optimization with Random Forest,PSORF)which is combination df random forest classifier and particle swarm optimization algorithm,the algorithm not only balances the virtual machine CPU utilization and memory utilization,will also make the execution time of the total tasks of the virtual machine as the optimized goal,which to balance the utilization of virtual machine resources while reducing the total makespan.Simulation results show that PSORF can effectively solve the load balancing problem of virtual machines.