Can MPI communication runtimes be designed to be energy-aware? Can energy be saved during MPI calls without a loss in performance? These are two questions briefly examined by Dhabaleswar K. (DK) Panda, of The Ohio State University, in a blog post on the Top500 website today (Designing Energy-Aware MPI Communication Library: Opportunities and Challenges).
Power, of course, is a huge issue in the push towards exascale systems. Panda, a professor and university distinguished scholar of computer science and engineering at OSU, points out the dominant current approaches to power management “treat the communication runtimes as a black-box with regard to energy consumption. Several of these techniques use DVFS to reduce the energy consumption of the communication phase of an application. However, such coarse-grain approaches lead to inefficient communication performance and hence, increase the total execution time of the application.”
Citing a student paper[i] from Ohio State University and Pacific Northwest National Laboratory, Panda suggests that power management might also be addressed by designing energy-aware MPI communication runtimes. The authors proposed a set of designs that exploit the slack in MPI calls to save energy by applying a lower energy lever, using DVFS and/or core-idling.
In the blog entry, Panda wrote, “For instance, with the Graph500 application kernel, it was demonstrated that the MPI runtime can achieve 41 percent energy savings with minimal impact on performance (less than 4 percent) using 2,048 processes.”
Here is a link to the full blog post: http://www.top500.org/blog/designing-energy-aware-mpi-communication-library-opportunities-and-challenges/
[i] A Case for Application-Oblivious Energy-Efficient MPI Runtime, Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis , Nov 2015 (Best Student Paper Finalist)