Preemptive Scheduling For Multiple DAGs Worklfow In Cloud Computing

In the research of cloud workflow technology application, a lot of progress hasbeen made, especially for the single workflow. And a larger number of schedulingstrategies were proposed at the target of the time, cost, safety, reliability. But they arestatic strategies. Namely, the communication cost, the computational cost and thecapability of the resource are certain. The assumption has a distance with the fact. Sothey can’t be directly applied to multiple workflow scheduling. The problem ofdynamic scheduling workflows and performance optimization are becoming more andmore important to satisfy the QoS demand for different users. What’s more, the natureof dynamic scheduling has the incomparable superiority. It is not only applicable toworkflow scheduling, but also can give attention to the scheduling performance.In the dissertation, we analyzed the research background, basic concept and thekey technology in the cloud computing such as the system structure and theprogramming model. The progress is described in detail, including the modern cloudcomputing platforms with the scheduling strategy. Then three layers multiple DAGdynamic scheduling model is proposed to solve the fairness and the utilization rateproblems. In the algorithm, priority division is in the light of the different QoSdemand of DAG, and the high priority DAG give preference to the resources. Thesame priority DAGs can preempt the resources according to the Slowdown with theheuristic information, improving the fairness of the workflow scheduling. In thecircumstances of high priority DAGs give preference to resources, the resourceutilization rate must be considered. Modified Backfill algorithm is introduced basedon threshold, improving the resources utilization rate with the guarantee of highpriority job in preference to resources. With the study of the fair scheduling algorithmin Hadoop, the model and preemption strategy are optimized and the improved FSPSalgorithm is proposed. At last the algorithm of scheduling policy is described. Weimplement the algorithm and the results show that the workflow scheduling more fairness and robustness. It works well in the guarantee high-priority work inpreference to resources and improving the utilization rate of resources at the sametime.