COUNTDOWN - three, two, one, low power! A Run-time Library for Energy Saving in MPI Communication Primitives
share
Power consumption is a looming treat in today's computing progress. In scientific computing, a significant amount of power is spent in the communication and synchronization-related idle times. However, due to the time scale at which communication happens, transitioning in low power states during communication's idle times may introduce significant overheads in the scientific application. In this paper we present COUNTDOWN, a methodology and a tool for identifying and automatically reducing the power consumption of the computing elements during communication and synchronization primitives filtering out phases which would detriment the time to solution of the application. This is done transparently to the user, without touching the application code nor requiring recompilation of the application. We tested our methodology in a production Tier-0 system, with the NAS benchmarks COUNTDOWN saves between the 6 time-to-solution penalty lower than the 5 --- Quantum ESPRESSO (QE) --- in a 3.5K cores run COUNTDOWN saves the 22.36 energy consumption with a performance penalty of 2.88 increases to the 37.74 application is executed by a non-expert users.
READ FULL TEXT