BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20151119T220000Z
DTEND:20151119T223000Z
LOCATION:19AB
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: As we approach exascale systems, power is turning from an optimization=0Agoal to a critical constraint. With power bounds imposed by both=0Astakeholders and the limitations of existing infrastructure, we need                                                                                           =0Anew techniques that extract maximum performance from limited power. In=0Athis paper, we find the theoretical upper bound of computational=0Aperformance on a per-application basis in hybrid MPI + OpenMP=0Aapplications.  We use a linear programming formulation to optimize=0Aapplication schedules under various power constraints, where a=0Aschedule consists of a DVFS state and number of OpenMP threads. We=0Aalso provide a mixed integer-linear formulation and show that the=0Aresulting schedules closely match schedules from the LP=0Aformulation. Across applications, we use our LP-derived upper bounds=0Ato show that current approaches trail optimal, power-constrained=0Aperformance by up to 41.1%. Our LP formulation provides future=0Aoptimization approaches with a quantitative optimization target.
SUMMARY:Finding the Limits of Power-Constrained Application Performance
PRIORITY:3
END:VEVENT
END:VCALENDAR