Research On Dynamic Scheduling Model Of Grid Service Workflow

Grid service workflow is a queue of member services with time constraints to achieve a specific goal. In the grid workflow, scheduling is an important issue. It not only affects the execution and efficiency of grid workflow, but also revolves the resource management.This thesis first gives the formal description of grid service workflow, and it introduces GSWL (Grid Service Workflow Language) to describe the workflow application for Information Service Grid. Then it proposes a hierarchical structure of the workflow system based on OGSA (Open Grid Service Architecture). After that, it presents a resource performance prediction model which can predict grid resource's using conditions and that could help the scheduler to select the best resource. Furthermore it designs and realizes a dynamic grid service workflow schedule model which considers the parallel among member services in a workflow by the use of ready queue and parallel queue. Based on this, it gives a dynamic grid service workflow parallel scheduling algorithm-PGSWA. Besides, the schedule model not only takes the parallelism between the services into account, but also considers the parallelism of an individual service. Combining the principles of self-scheduling, it designs grid service parallel self-scheduling algorithm (GSPSS) to implement the parallelism in a member service. Finally, it realizes the CRF-based segmentation workflow application based on that scheduling model and the experimental results show that our model can shorten the execution time of word segmentation program effectively and improve the grid resources utilization.