Improved Task Scheduling for Parallel Simulations
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
The objective of this investigation is to design, analyze, and validate the generation of optimal schedules for simulation systems. Improved performance in simulation execution times can greatly improve the return rate of information provided by such simulations resulting in reduced development costs of future computerelectronic systems. Optimal schedule generation of precedence-constrained task systems including iterative feedback systems such as VHDL or war gaming simulations for execution on a parallel computer is known to be N P-hard. Efficiently parallelizing such problems takes full advantage of present computer technology to achieve a significant reduction in the search times required. Unfortunately, the extreme combinatoric explosion of possible task assignments to processors creates an exponential search space prohibitive on any computer for search algorithms which maintain more than one branch of the search graph at any one time. This work develops various parallel modified backtracking MBT search algorithms for execution on an iPSC2 hypercube that bound the space requirements and produce an optimally minimum schedule with linear speed-up. The parallel MBT search algorithm is validated using various feedback task simulation systems which are scheduled for execution on an iPSC2 hypercube. The search time, size of the enumerated search space, and communications overhead required to ensure efficient utilization during the parallel search process are analyzed. The various applications indicated appreciable improvement in performance using this method.
- Computer Programming and Software
- Computer Hardware