Electrical energy aware parallel computing with MPI and CUDA
Aleksandr Igumenov, Julius Žilinskas · 2013 Eighth International Conference on P2P, Parallel, Grid, Cloud and Internet Computing · 2013
The object of research of the article is environment-friendly technologies that can minimize electrical energy consumption of parallel and distributed computers. The study revealed that even without computations computer clusters consume considerable amount of electrical power. A technique to save electrical energy is proposed to automatically shut down compute nodes when they are not needed for computation, leaving only network cards on waiting for the tasks. When the tasks are assigned, electrical power consumption increases, but not evenly for each core - each next core increases consumption of power less than the previous one. Therefore for efficient use of electricity computer's cores must be loaded as much as possible, rather than parallel tasks divided among several computers. Graphics processing units consume a lot of electrical power. However their high computational performance makes computing electrical energy efficient when they are used with care.