Execution Plan Generation Of Hybrid Parallel Scientific Workflow In Cloud Environment

In recent years,scientific workflow has been widely used in many fields,especially in business and scientific research.Scientific workflows have the characteristics of large data volume,long execution time,and complicated calculation process.The execution of scientific workflows may have multiple parallel modes,such as data parallelism,task parallelism,and pipeline parallelism,which makes the execution of workflow more complex.It has become undesirable to execute scientific workflows using a single physical computing environment.In theory,the cloud computing environment can provide unlimited storage and computing capacities,enabling users to dynamically and flexibly access computing resources,which makes it an ideal choice for scientific workflow execution.Therefore,it is important to improve the execution efficiency of hybrid parallel scientific workflow in cloud environment,reduce its execution cost and generate its reasonable execution plans.Focusing on the above problems,this thesis explores the execution plan generation of hybrid parallel scientific workflow with the aim of improving its execution efficiency and reducing the execution cost.In order to reduce the cost of scientific workflow execution in cloud environment,an approach to optimizing the execution plans of scientific workflows in cloud is proposed.The Monkey Group Algorithm is introduced to facilitate the intra-level and inter-level optimization of the current execution plan under the constraint of deadline.Through the clustering of tasks in the same level and adjusting the level a task belonging,the number of tasks respective to different levels in the execution plan can be balanced.Therefore,it can reduce the resource idling and the accumulated delays of tasks.Experiments show that compared with the BTS algorithm and the SPSWVC algorithm,the proposed method consumes less resource and significantly reduces the total delay time of tasks.As to the execution plan generation of hybrid parallel scientific workflow,topological sorting is used to partition three kinds of possible task layers.For each layer,the execution time of the same layer tasks is roughly the same,and the task logic aggregation problem is described as a set partition problem,and the difference evolution algorithm is used for task logic aggregation.The tasks that can be sliced are equally fragmented according to the sum of the execution times of each subset.Under the global deadline constraint of hybrid parallel workflow,the waiting time of task execution is reduced.So,a reasonable execution plan of hybrid parallel scientific workflow is obtained.Experiments show that the proposed method consumes less resource and significantly reduces the total delay time of workflow tasks,and largely utilize the idle time of resources and reduce the cost of execution.