Scheduling algorithms for multi-core and GPU

Abstract

Multiprocessor systems are now widely used as solution to the performance limitation and power problem currently limiting single processor systems. However to fully exploit the job parallelism on these systems, its resources(processors) must be assigned efficiently. In addition to improving the program's performance, the power consumption must also be minimized. In this project, adaptive scheduling and algorithms to reduce the power consumption will be studied. The algorithms are evaluated based on the response time and power consumption. Equiparttioning (EQ) divide the number of processors equally to every job. Dynamic Equiparttioning (DEQ) operates the same way as EQ with a slight difference. Jobs are never allocated more than what was requested in DEQ. Adaptive Greedy Dynamic Equipartitioning(AGDEQ) operates the same way as DEQ except that the next cpu allocation is done by using the utilization rate of the previous task execution. In terms of response time, AGDEQ has the lowest response time. This is followed by DEQ and lastly EQ. In terms of power consumption, while Geometric power distribution can improve the response time, it also consumes the most power. The scheme that has the lowest power consumed is equal power distribution among processors.