Clusters are rapidly becoming the low-cost and the standard platforms for parallel computing. This paper focuses on task scheduling problem in a cluster system that is the most challenging in parallel computing. We present a new task scheduling system in which the target machine is a dynamic larger-scale parallel cluster (DLPC) and the parallel program is a special task synchronization graph (TSG). The TSG model is introduced to solve synchronous cooperating working problem among coarse-grained tasks in message passing system. We describe a scheduling algorithm (STF) that maps TSG onto the DLPC cluster. Our study indicates that STF algorithm can reduce the completion time of the parallel program.
展开▼
