Performance evaluation of scheduling precedence-constrained computations on message-passing systems

摘要

Using knowledge on computation, communication, and multiprocessor topology, a class of global priority-based scheduling heuristics, called generalized list scheduling (GLS) is proposed. Task-priority is defined as the completion time of the task following backward scheduling the computation over the multiprocessor by using the best local heuristic. GLS scheduling consists of using the task-priority in forward, graph-driven scheduling. Evaluation of local (ETF) and GLS heuristics is carried out by altering over the communication, parallelism, and system topology. Analysis shows that local heuristics rely on locally maximizing the efficiency and gives acceptable solutions only when the parallelism is large enough to cover the communication (bounded speedup). GLS scheduling outperforms the local approaches versus change in parallelism, communication, and network topology. The time complexity of GLS heuristics is O(pn/sup 2/), where p and n are the number of processors and that of the tasks, respectively.