On-Line Optimization Techniques for Task Scheduling in Multiprocessor Systems

Efficiently scheduling parallel tasks on to the processors of a multiprocessor system is critical to achieving high performance. Given perfect information at compile-time, a static scheduling strategy can produce an assignment of tasks to processors that ideally balances the load among the processors while minimizing the run-time scheduling overhead and the average memory referencing delay. Since perfect information is seldom available, however, dynamic scheduling strategies distribute the task assignment function to the processors by having idle processors allocate work to themselves from a shared queue. While this approach can improve the load balancing compared to static scheduling, the time required to access the shared work queue adds directly to the overall execution time. In this paper, we introduce a class of algorithms (i.e. Self-Adjusting Dynamic Scheduling (SADS)) that undertake an on-line optimization strategy to dynamically compute partial schedules based on the loads of the other processors and the memory locality (affinity) of the tasks and the processors. We also introduce a general framework for classifying and characterizing different scheduling strategies. The paper provides analytical and empirical analyses of the SADS family of scheduling algorithms. Our results show that the SADS algorithms outperform existing dynamic scheduling algorithms by performing optimization techniques while explicitly controlling the scheduling time. Ke y Words: dynamic scheduling, scheduling costs, load balancing, locality management. Name email phone fax Babak Hamidzadeh [email protected] 852-2358-7011 852-2358-1477 David J. Lilja [email protected] 612-625-5007 612-625-4583