A Grid DAG Scheduling Algorithm Based on Fuzzy Clustering

Abstract

Focusing on the problem of task scheduling under large-scale, heterogeneous and dynamic environments in grid computing, a heuristic algorithm based on fuzzy clustering is presented. Many previous scheduling algorithms need to search and compare every processing cell in the target system in order to choose a suitable one for a task. Though those methods can get an approving Make-span, undoubtedly, it would increase the entire runtime. A group of features, which describe the synthetic performance of processing cells in the target system, are defined in this paper. With these defined features, the target system, also called processing cell network, is pretreated by fuzzy clustering method in order to realize the reasonable clustering of processor network. In the scheduling stage, the cluster with better synthetic performance will be chosen first. There is no need to search every processing cell in the target system at every scheduling step. Therefore, it largely reduces the cost on choosing which processing cell to execute the current task. The design of the ready task's priority considers not only the influence that comes from the executing of nodes on critical path, but also the influence induced by heterogeneous resource, on which the task will be scheduled. In the last part, the algorithm's performance is analyzed and compared with other algorithms, and the test results show that the bigger the target system is, the better performance the algorithm shows.