Research On The Cloud Task Scheduling Based On Evolutionary Multi-tasking Algorithm

Cloud computing has brought convenience to users and enterprises.However,as the number of tasks submitted by users,the number of virtual machines,and the need for heterogeneity continue to increase,leading to the increasing complexity of cloud size.Evolutionary algorithms require a large number of computational resources as well as a runtime in solving to obtain a feasible set of scheduling solutions.In this paper,the evolutionary multitasking algorithm is used as the research context to redescribe the cloud task scheduling model into the form of multitasking optimization,solve it using the evolutionary multitasking algorithm,and design an optimization algorithm with better performance.(1)For the multi-objective cloud task scheduling problem,firstly,the problem is decomposed into a set of simple multi-objective optimization sub-problems using weighted aggregation of objective function subsets based on the idea of decomposition.Secondly,multi-factor optimization is applied to this scheduling problem,and an MFO-D algorithm combining task scheduling features with multi-objective multi-factor optimization is proposed to construct an auxiliary optimization task.Finally,a dynamic migration strategy is designed to dynamically control the magnitude of knowledge migration probability in the MFO-D algorithm by measuring the similarity between the auxiliary tasks from the overlap degree of the problem.(2)The complexity of computational resources increases because a large set of tasks uploaded by users is received on the cloud.A large-scale multi-objective cloud task scheduling model is proposed to simulate the cloud scheduling problem in realistic scenarios.The appropriate virtual machine resources are scheduled by considering the user-submitted task requirements while considering the task execution time and task execution cost.Second,a multi-factor-based NSGA-III algorithm is proposed.The experimental results demonstrate that the method can obtain the best scheduling solution while maintaining good objective function results compared with other optimization algorithms.(3)To provide a more efficient solution algorithm for the constructed problem model,an adaptive multi-group multi-objective algorithm A-MPMO framework is proposed in this paper.The current algorithms produce very different optimization effects when dealing with complex models due to different parameter settings.Therefore,eliminating the objective function result diameters is a meaningful research problem.First,A-MPMO divides the population into multiple subpopulations to expand the search range,and they are updated iteratively using operators with different genetic parameters,respectively.Second,based on multiple populations,the subpopulations compete with each other for limited computational resources,i.e.,the size of each subpopulation is adaptively adjusted according to its contribution to problem-solving.Finally,the set of subpopulations that is most suitable for solving the problem will be selected.The experimental results demonstrate the better performance of A-MPMO compared with other multi-objective optimization algorithms on benchmark test suites DTLZ,ZDT,and UF.The A-MPMO algorithm is used to solve the problem model presented above,and the results show that A-MPMO outperforms all other comparative algorithms,further proving the effectiveness of the algorithm.