Research Of The Cost-driven Scheduling Of Cloud Workflows Based On EC2

With the developing of scientific research, scientific computing usually need to carry out with the help of workflow technology in large-scale distributed systems. Cloud computing is a new distributed computing model by using virtualization technologies with some advantages like on-demand supply resources, high reliability and high extensibility. In order to overcome the geographical restrictions and save the software and hardware resources, we can run the workflow on the cloud computing platform. However, when we adjust a workflow task to different rsources, it can produce different rents. To reduce the cost of executing workflow, this paper will make the execution time and the cost of workflow as the QoS parameters.The thesis analyzes the elastic characteristics of cloud in depth.The workflow is devided into non-preemptive and preemptive according to wheter the resource is contention. We establish the corresponding Elastic Compute Cloud scheduling model and put forward two kinds of non-preemptive workflow scheduling algorithm, and a preemptive workflow scheduling algorithm.For non-preemptive scheduling workflow, because there is no resource contention of multiple workflows, this problem can be converted to a single workflow scheduling one. There is only one global deadline for single workflow scheduling, and different pricing model of scheduling algorithm is put forward combined with critical path. Based on the time billing mode, the billing cycle will affect the cost of the workflow. Part of the critical path scheduling algorithm is put forward, and part of the critical path of workflow is scheduled through the depth first search. In the task-based billing model, the impact of the cost is execution time. We can propose iterative critical path scheduling algorithm and use dynamic programming on the critical path of the workflow scheduling.For preemptive scheduling workflow, to solve the problem of multi-process fairness issues, to improve utilization rate of resources, and to meet the different QoS requiremnts of different priority workflow, workflow scheduling algorithm based on priority is proposed. The algorithm will prioritize workflow according to the QoS demand, and adopt the principle of high-priority job priority possess high performance resources, so that each workflow can be completed before the deadline.The average high-priority workflow execution time is short, while reducing the cost of resource rent.Finally, the workflow scheduling experimentizes on simulation based on CloudSim simulation platform and compare the cost, the running time and scheduling fairness index with algorithm based on the Dealine-MDP respectively. Experimental results show that the proposed cost-driven cloud workflow scheduling algorithms are able to reduce the cost at a relatively lower time complexity effectively, ensure good scheduling fairness, and improve service quality.