Show Entry in Conferences and Workshops

You may return to the Main Menu when done.

Showing 1 entry


[sc15]
Daniel A. Ellsworth, Allen D. Malony, Barry Rountree, Martin Schulz: Dynamic power sharing for higher job throughput. SC 2015: 80:1-80:11

Keywords: RAPL; hardware over-provisioning; HPC; power bound

Current trends for high-performance systems are leading towards hardware overprovisioning where it is no longer possible to run all components at peak power without exceeding a system- or facility-wide power bound. The standard practice of static power scheduling is likely to lead to inefficiencies with over- and under-provisioning of power to components at runtime. In this paper we investigate the performance and scalability of an application agnostic runtime power scheduler (POWsched) that is capable of enforcing a system-wide power limit. Our experimental results show POWsched is robust, has negligible overhead, and can take advantage of opportunities to shift wasted power to more power-intensive applications, improving overall workload runtime by as much as 14% without job scheduler integration or application specific profiling. In addition, we conduct scalability studies to determine POWsched’s overhead for large node counts. Lastly, we contribute a model and simulator (POWsim) for investigating dynamic power scheduling behavior and enforcement at scale.

URLs:

Modified: Fri Jan 20 14:35:44 2017
Created: Fri Jan 20 14:31:45 2017


Current Collection: Conferences and Workshops
[ Menu | List | Show | About ]

Return to the ParaDucks Research Group Publications page.