A Two-Stage Estimation Of Distribution Algorithm For Cloud Workflow Scheduling

With the rapid development of network technology,cloud computing mode has been involved in all aspects of real life,and more and more people pay attention to cloud data center which provides a computing platform for massive data processing.At present,researchers generally pay attention to workflow scheduling under cloud computing.For a task that needs to be calculated,it has become an urgent practical problem to efficiently and timesaving task scheduling and resource allocation.Due to the rapid growth of the amount of data to be calculated and the diversity of computing platform resources,the cloud workflow scheduling platform has some problems in the scheduling process,such as low resource utilization and long data processing time.Therefore,this paper mainly studies the cloud workflow scheduling process,analyzes and models the problems studied,and proposes a twostage estimation of distribution algorithm,which is proved to be effective by experiments.This paper first introduces the background of cloud workflow scheduling optimization,then introduces the relevant overview,and gives the description of the problem and scheduling optimization model.Secondly,A two-stage estimation of distribution algorithm(TSEDA)is proposed to solve the problem that the current heuristic algorithm is dependent on specific problems and the meta-heuristic method has the problems of incomplete search space or low search efficiency in complete space.The algorithm adopts integer encoding,and each solution in the search space has a corresponding encoding mode,so it can realize the global search.Before the algorithm starts the iterative search,an individual generated by the heterogeneous earliest finish time(HEFT)is seeded into the population,and the search starting point of the solution is improved.In the process of evolution,the task scheduling order probability model and resource allocation probability model are updated respectively,so that the algorithm constantly searches for the optimal solution.In addition,a two-stage coevolution strategy is designed.In the first stage,the task scheduling sequence list of individuals based on probability model sampling and the heuristic strategy based on the earliest finish time are used to generate the resource allocation list and decode individuals,so as to accelerate the convergence speed of the algorithm and make the algorithm converge near the optimal solution as soon as possible.In the second stage,the task scheduling order list and resource allocation list of individuals are generated based on probability model sampling.The insertion mode decoding method is used to expand the search combined with two improved strategies of backward decoding improvement(FBI)and load balancing improvement(LDI),so as to improve the neighborhood/local search ability of the algorithm.Finally,the effectiveness of the TSEDA algorithm proposed in this paper is verified.The algorithm is implemented in C++ programming language,and compared with HEFT,HGA,LWSGA and CGA.The experimental results show that compared with other existing algorithms,the two-stage estimation of distribution algorithm performs better in the efficiency of searching solutions and can find better solutions at the same time.