Execution Plan Generation Of The Scientific Workflow In Cloud

Scientific workflow is a novel application paradigm emerging in recent years,which is usually either data-intensive and computing-intensive.Therefore,the traditional computing environment can not act well to provide the computing recources for scientific workflows.The emergence of cloud computing gives a good choice for the resources provision of the execution of scientific workflows,because it can provide infinite computing and storage capabilities theoritically as an extension of existing distributed computing,parallel computing and grid computing techniques.Therefore,it can meet the special needs of scientific workflows.The execution of scientific workflows in cloud environment requires to rent infrastructure resources and pay for its use.So,it is important to estimate the number of processors required for a scientific workflow and to generate a reasonable scientific workflow before hand.The rational allocation of resources and the proper mapping of workflow tasks to the computing resources is the basic for a scientific workflow to run efficiently in the cloud environment.It has a big impact on the execution efficiency and execution cost of the workflow.The essential problems under such a process can be called the generation of execution plan.To be noted that determining the minimum number and types of processors is the most key issues to generate the execution plan.Therefore,this paper presents an execution plan generation strategy based on the packing principle firstly.It intends to reduce the number of processors used.The tasks are partitioned and sorted in each partition according to whether it is on the key path.Then,the tasks are mapped to the processors according to the best adaptation rules commonly used in the Packing Problem under the deadline constraints of the scientific workflow.Secondly,considering the execution of the scientific workflow is dynamic and the computing resources released by a task can be reused by other,an approach of the execution plan optimization are proposed based on the task scheduling and replication.It can improve the rate of the resource utilization and save the cost of exectution of a scientific workflow.Experiments and simulations show that the policy and approach proposed can support to generate the execution plan of a scientific workflow,especially,address the mapping of tasks to processors,determine the minimum number of processor necessary.With the dynamic optimization of the execution plan,it can improve the utilization of the processor,and ultimately reduce the cost of the execution of a scientific workflow.