Research On Dynamic Scheduling Algorithm Based On Performance-Driven For Grid Workflow

Grid technology is proposed to share resources for research collaboration, with the rapid development of grid, the application of grid is becoming more and more complex, In order to build, implement, monitor and manage the grid applications, workflow technology is introduced into the grid. Grid workflow becomes the key factor of grid technology, and scheduling is the critical problem to be solved in grid workflow, proper scheduling can have significant impact on the performance of workflow system, this thesis focuses on the dynamic scheduling problem based on performance-Driven.Firstly, this thesis introduces the conception of grid, grid workflow and workflow scheduling; expounds the importance of scheduling in the research of grid workflow technology. Then a grid workflow management based on workflow reference is proposed, also we summarize the modules of grid workflow management in detail.Then, we analyze the existing technology of grid workflow scheduling algorithm; find out the respective defects. On the basis of DCP-G algorithm which based on dynamic critical path, we elaborate the concept of idle time. The idle time on resources will reduce the utilization of grid resources. In this thesis, dynamic critical path-data prefetching algorithm(DCP-DPA) is proposed, which can effectively reduce the idle time, make scheduling more efficient. When using priority scheduling algorithm in communication-intensive workflow, it will occur the phenomenon of activities aggregation, and the activity duplication algorithm can make up that defect by reconstruct workflow model.At last, on the basis of GridSim platform, we extend the workflow model and implement the scheduling algorithm. With different experimental models and workflow parameters, we verify that our algorithm can be more efficient compared with the DCP-G algorithm.Through the experiment, priority scheduling algorithm has been verified its limitation in scheduling communication-intensive workflow, and the redundant activity algorithm can effectively solve the problem.