Research On Cloud Workflow Scheduling Based On Genetic Algorithm Of ARIMA Prediction Model

With the development of cloud computing,many large-scale and complex scientific applications have migrated to the cloud platform and used computing resources on the cloud to complete computing tasks.Unlike traditional grid computing,cloud computing provides a customizable infrastructure that encapsulates computing,storage,and other resources in different geographic locations into cloud services,and delivers them in a “pay-as-you-go,on-demand” model.Supply resources.The workflow system on Infrastructure-as-a-Service cloud defines abstract and complex processes of scientific applications and utilizes resources such as computing and storage to complete these tasks of scientific computing and big data storage.In this way,how cloud service providers to allocate computing resources,i.e.,minimizing the total cost under the Service-Level-Agreement constraint conditions,has become an urgent problem.For this problem,academic community has conducted extensive research.However,there is a common defect in related research,i.e.,the performance of a virtual machine is assumed to be constant or changed within a specific range,and then a scheduling scheme is proposed.Nevertheless,the assumption can lead to potential Service-Level-Agreement defaults or waste of computing resources.Virtual machines are a collection of computing resources after virtualization,which is affected by various factors like resource pool changes and network connections between cloud nodes during operation,and its Quality of Service is dynamically fluctuating.For above problems,this essay study on the improved schemes.This paper innovatively considers the dynamics of performance fluctuations on virtual machines,and considers that its performance fluctuations change with time,under the deadline constraint condition,and the minimum cost is our scheduling target,then an prediction-based genetic algorithm is proposed.The method first predicts the execution time of the tasks on different virtual machines by ARIMA prediction model.And then puts them as data inputs for the cloud workflow scheduling model established in this paper,which obtains the cost of the different virtual machines running workflows and the estimated completion time.After genetic operations such as chromosome coding,population initialization,selection,crossover,and mutation,the fitness evaluation based on cost and time is applied to generate an optimal scheduling scheme after repeated iterations.In order to verify the feasibility and effectiveness of the proposed method,this paper implements different types of cloud workflow tasks on three different cloud platforms,namely Huawei Cloud,Tencent Cloud and Amazon EC2 Cloud.The simulation experiments is based on Matlab platform,and the non-predictive genetic algorithm,non-predictive particle swarm optimization and ARIMA-based particle swarm optimization algorithms are compared.Finally,the experimental results show that the proposed method outperforms other algorithms in terms of SLA violation rate,execution time and execution cost.The scheduling goal of minimizing cost under the condition of satisfying the SLA deadline constraint is realized.