Stochastic scheduling of parallel processors
Selected topics of interest from and area of parallel processing systems are investigated. Problems concern specifically an optimal scheduling of jobs subject to a dependency structure, an analysis of the performance of a heuristic assignment schedule in a multiserver system of many competing queues, and the optimal service rate control of a parallel processing system. In general, multi-tasking leads to a stochastic scheduling problem in which n jobs subject to precedence constraints are to be processed on m processors. Of particular interest are intree forms of the precedence constraints and i.i.d. job processing times. Using an optimal stochastic control formulation, it is shown, under some conditions on the distributions, that HLF (Highest Levels First) policies and HLF combined with LERPT (Longest Expected Remaining Processing Time) within each level minimize expected makespan for nonpreemptive and preemptive scheduling, respectively, when m = 2. The relative performance of HLF heuristics are investigated for a model in which the job execution times are i.i.d. with an exponential distribution. Many situations in resource sharing environments can be modeled as a multi-server system of many competing queues.
- Research Organization:
- California Univ., Berkeley (USA)
- OSTI ID:
- 6891504
- Country of Publication:
- United States
- Language:
- English
Similar Records
Tighter bounds for LPT scheduling on uniform processors
Approximation techniques for average completion time scheduling