Research On Workflow Task Scheduling Method For Performance Fluctuating Cloud Infrastructure

Cloud computing is a new computing resource service model based on the network.It can provide flexible computing resources and flexible public services,allowing users to customize different infrastructure to obtain resources according to demand,and adopting the charging model of pay on demand,providing users with an efficient and economic computing platform.With the continuous development of cloud computing,complex large-scale scientific applications and economic problems are often modeled as workflows and migrated to the cloud platform.The computing requirements and data scale of these workflow applications are constantly increasing,which makes the relevant research on efficient scheduling of workflow tasks in cloud computing environment receive extensive attention.The workflow management system of cloud computing can abstractly define the complex process of scientific application,and use computing resources to complete tasks such as scientific computing and data storage.However,it is different from the traditional high-performance computing platform in terms of resource provision mode and payment mode.At the same time,its services are also affected by real-time performance changes and fluctuations,so it is difficult to ensure the efficiency of cloud based workflow Cost effectiveness and service quality.These are the important challenges of current cloud workflow scheduling.How to efficiently allocate computing resources,optimize economic costs,and ensure service quality has become an urgent problem.For this problem,the academia has carried out extensive research on workflow scheduling in cloud environment.In the existing research,many methods still have some limitations,such as assuming that the service performance of virtual machine is constant or has a specific distribution,and generating a scheduling scheme based on it.Some scheduling methods describe it as a static optimization problem,ignoring the actual execution process of workflow.These assumptions may lead to miscalculation of computing resources,waste of resources and additional costs,or even violate the service level agreement,resulting in serious deterioration of service quality and serious impact on user experience.This thesis studies the above limitations,innovatively considers that the computing process of scientific workflow supported by cloud infrastructure is volatile and dynamic with time,and proposes a new scheduling strategy to schedule the cloud workflow with the deadline as the constraint and the objective of minimizing the economic cost.Firstly,the ARIMA model is used to predict the performance of cloud resources,which is used as the parameter input of scheduling algorithm.Then,the static scheduling algorithm based on genetic algorithm and the dynamic scheduling algorithm based on critical path algorithm are designed.The scheduling strategy of combining static scheduling and dynamic scheduling is adopted to avoid performance fluctuation and prediction error The impact of poor performance and implementation bias.Finally,in order to verify the feasibility and effectiveness of the proposed method,this thesis constructs different types of workflow tasks on three different third-party commercial cloud platforms:Huawei cloud,Tencent cloud and Baidu cloud.The simulation results show that the proposed method is superior to the traditional method in terms of execution cost,completion time and violation rate of Service-Level-Agreement.