Software Based Automatic Self-Adjusting Threads (ASAT) for Parallel Processors

A number of commercial companies are now attempting to develop production quality parallel processing systems. These systems are often targeted to take over the role of Vector/Parallel supercomputers. While these systems have excellent raw compute power for their cost, they lag behind traditional supercomputers in a number of important areas. One important area is the ability to deal with wide variations in system load while maintaining high utilization. This paper discusses the benefits of "Automatic Self-Adjusting Parallel Processing" as used in traditional supercomputers and how this technique can be applied to parallel processing systems. The paper also proposes a solution to the problem which allows these new processors to adapt to changing load conditions through software. This software based "Automatic SelfAdjusting Threads" or ASAT is discussed in the paper and its performance is compared to the hardware based solutions. Keywords: Parallel processing, dynamic load balancing, co..