Multiprocessor scheduling algorithm utilizing linear clustering of directed acyclic graphs

초록

The multiprocessor scheduling is the method to allocate modules(or tasks) with precedence relationship onto the processors in such a way that the parallel time(completion time of a program in multiprocessor system) is minimized. In this paper, we propose a scheduling heuristic which is based on linear clustering scheme and sues only local information of a DAG(Directed Acyclic Graph). Proposed algorithm gives an optimal scheduling result for a coarse grain DAG with only fork or join nodes and a greatly reduced parallel time for general DAGs. Simulation results showed that proposed algorithm has better scheduling result and less algorithm execution time than any of compared algorithms