Task Scheduling Algorithm For Cloud Computing Based On Improved Differential Evolution

In recent years,due to the advantages of strong computability,scalability,low expense,flexible resource management,and quick deployment,cloud computing has attracted great interests from both the industry and the academica.Task scheduling is one of the basic problem on cloud computing,which according the needs of tasks,allocates different tasks to the right resources nodes using the proper policies in the cloud.In the cloud computing systems,task scheduling is the way that maps tasks to data clusters combined with large number of heterogeneous nodes,all tasks must be the allocated to compatible computing nodes considering performance of each node.This has proved to be a non-deterministic polynomial.The research of the task scheduling and resource allocation is few,and the existing task scheduling algorithms for cloud computing usually lay their attention on the pursuit of the shortest completion time,however,they are not well to take into account the cost of all the tasks for the pursuit of the shortest completion time.Because the cost of all the tasks is a important factor which can not be ignored.Different computing resources in cloud platform have different costs,the use of excellent computing resources is higher,while the use of common resources is lower.The cloud users usually make their choices based on comprehensive consideration of the economic budget as well as the wait time.To solve the problem,the main research work and contributions of the thesis are as follows:(1)We propose an Improving Differential Evaluation(IDE)algorithm,which can solve the limitations exist in the Differential Evaluation(DE)algorithm,such as slow convergence,difficult parameter setting and Local optimum.The idea of our IDE algorithm is described as follows.First,since the basic crossover operator of the DE is able to generate two offspring individuals,which can retain the most effective genetic information with the minimum number of simulations.Second,as the time used to evaluate the scheduling scheme is very long,it is difficult get the value of the control parameters rely on large number of experiments.we improve the scaling factor F and the crossover probability CR,so that the control parameters can be automatically adjusted with the evolution of the algorithm.(2)In consideration of users and data centers of cloud computing,we propose a cloud computing scheduling model based on the multi-objetive optimization scheduling model,which selects the running time and costs of cloud computing as the goal.Compared with the DE algorithm,the task scheduling in cloud computing constrained by the cpu,memory,tolerance time and so on.Therefore,we first process these constrains with the rule-based processing mechanism,then research on open source cloud computing simulator CloudSim,and apply the DE and IDE to our multi-objective optimization scheduling model.The simulation results show that our IDE algorithm perform better in the aspects of makespan and cost than the DE.