Abstract
Power consumption is one of the most important challenges facing the exascale computing community. Current technology trends will not provide a reasonable power budget for the next generation of exascale supercomputers in near future. Finding new ways to improve power efficiency will become a major challenge to exascale computing researchers.
This study was motivated by the idea that some algorithms may have power advantages that need to be identified for exascale systems. In this work, we compared the power efficiency of a basic mergesort to the classically optimized 3-way partitioning quicksort. We measured the variations of average power between the two sorts that result from running different types of workloads for each algorithm. The study showed that empirical evidence suggests a generic mergesort has some power efficiency advantage over the 3-way partitioning quicksort.