Research On Many-objective Task Scheduling Strategy In Hybrid Cloud Environment

Nowadays,the growing variety and scale of digital business is driving the cloud computing infrastructure toward a more flexible,intelligent and scalable direction.As a form of cloud computing,hybrid cloud combines the features of different architectures and is able to give full play to the core advantages of cloud computing.However,the flexibility,heterogeneity,and complex and diverse user requirements of hybrid clouds also bring great challenges to the task scheduling problem.Therefore,this paper takes the hybrid cloud environment as the research background to open the connection between private and public clouds.A many-objective task scheduling model is constructed and solved by combining the diverse user requirements.(1)To effectively address the issues of business burst,application load expansion,privacy data protection,green scheduling and user diversification requirements in hybrid cloud environment,a many-objective task scheduling model for hybrid cloud environment is constructed in this paper.The model takes the amount of leased public cloud resources and task allocation strategy as decision variables,task completion time,user cost,cloud resource energy consumption and resource utilization as optimization objectives,and ensures user task data privacy as constraints.Simulation experiments are conducted for the model using the classical many-objective optimization algorithm to verify the effectiveness of the model.(2)Due to the unique decision variables in the task scheduling model increase the complexity of problem solving,and the traditional evolutionary operator cannot generate effective individuals for the model.Therefore,a symmetrical population region-division many-objective task scheduling algorithm is designed in this paper to generate symmetrical populations for the case that traditional region-division methods do not consider the subspace of unassigned individuals,thus enabling a more uniform distribution of individuals in the population.In addition,a crossover-mutation strategy and a deviation gene-adjustment strategy are designed according to the constraints of the hybrid cloud task scheduling model,and the evolutionary operator is improved to generate offspring individuals on the basis of guaranteed data privacy and resource expansion.The effectiveness of the algorithm on DTLZ,MaF standard test set and task scheduling model is verified by simulation experiments.(3)Considering that the performance of the symmetric population region division scheduling algorithm depends mainly on the definition of the reference information in the objective space region.For the problem types with irregular or unknown optimal frontier surface,excellent solution sets cannot be obtained.Therefore,a adaptive reference vector selection algorithm for many-objective task scheduling is proposed in this paper based on the algorithm in Chapter 2.By designing an adaptive reference vector selection and adjustment strategy,suitable individuals in each generation of population are selected as reference vectors,so that the distribution of reference vectors and the optimal frontier surface form a positive feedback in the iterative process to improve the performance of the algorithm.And combine the reference vector distribution to improve the offspring selection strategy to save the individuals with better convergence and diversity into the next generation population.Finally,the performance of the algorithm and other popular algorithms are compared on DTLZ,MaF test set and task scheduling model under the same evaluation index to verify the effectiveness of the algorithm.