| Scientific computing has been shifting from the centric structure to the distributed one during the past decade, and recently has extended to the grid computing. The goal of grid is to share resources, to support remote and parallel accesses to computing resources. Because of the characters of distributed, shared, self-resembling, dynamic, diverse, autonomous and manifold management of grid, the conventional workflow technology can't effectively solve the relevant problem of the grid environment. According to characteristics of grid resources and grid applications, the thesis studies the grid workflow scheduling algorithm. The details are as following:The basic concepts, character and application domains of the grid workflow are introduced; then the present situation and development trend of grid workflow scheduling are presented and analyzed in this paper. Workflow scheduling is one of the key issues in the workflow management. The existing schedulings emphasize particularly on various points, however, few of them really take account of the volatility of resource. For example, the grid resource is based on the assumption that the machines on the grid never break down or never present abnormal performance when some workflow tasks are running on them in those algorithms. In fact, the execution time of workflow is affected by'normal state'and'abnormal state'of grid resource occurring in successive turns. In order to produce a good schedule, this paper presented two stochastic algorithms in terms of finite-state continuous-time Markov process to obtain scheduling solutions for workflow tasks with selecting the lowest expenditure of scheme of resource system under the certain credit level of the reliability of resource combination on the critical path in DAG.Finally, with the help of Matlab, the emulation of grid resource scheduling with price requirements is realized. The experiment results show that the new scheduling algorithms presented in this paper is feasible. |
PDF Full Text Request