Research On Cost Optimization For Scheduling On Clouds

In the era of big data,cloud computing provides a more convenient computing environment for big data processing,enabling users to focus on the computing requirements without too much consideration of the computing environment.As cloud computing services prevail in commercial markets,new pricing policies are emerging where cloud providers resource provisioning based on the allocated configuration of virtual machines(VMs).Cloud service providers now tend to allow their customers to customize resources that meet their needs freely.For users,how to select appropriate virtual machine resource settings for tasks,to minimize the cost of computing while satisfying the quality of service(QoS),has become a new problem that puzzles them.This paper mainly explores this problem,our research goal is to optimize the makespan and cost of workflow.This paper describes the workflow scheduling model in cloud environment.According to previous work,workflow is modeled as directed acyclic graph(DAG)and three different pricing models are investigated in this paper.Based on new performance-based pricing models,aiming at optimizing makespan and cost of workflow,the research carried out in this paper is as follows.:(1)For cost-eff-icient with clear deadlines,an algorithm framework based on greedy idea is proposed,and three algorithms CFMax,CRR and CBT are extended from the framework.This framework uses HEFT to get the initial scheduling scheme.It iteratively select resources that meet the expectations of the algorithm and map them to the corresponding tasks.Then,it check the makespan of the workflow before updating the scheduling scheme to ensure that the deadline limit is met.The costs obtained by the three algorithms are all significantly lower than HEFT,and in most cases these algorithms are better than CSFS-Max.When the deadline is loose,CFMax can reduce the computing cost by 76%compared with HEFT.(2)For simultaneously optimize the makespan and cost of workflow,a multi-objective optimization algorithm for workflow scheduling named SABD is proposed based on a decomposition-based multi-objective evolutionary algorithm,MOEA/D.SABD uses CFMax's intermediate iteration process to construct the initial population,and alternately searches for new solutions by minimizing task computation cost and task execution time.In order to compare the effects of three decomposition methods on SABD algorithm,the hypervolume value is used to measure the optimal solution set.And among these three methods,Tchebycheff has better effect and more stable.