BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20151119T200000Z
DTEND:20151119T203000Z
LOCATION:18AB
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: The last decade has seen power consumption move from an afterthought to the=0Aforemost design constraint of new supercomputers.  Measuring the power of a=0Asupercomputer can be a daunting proposition and, as a result, many published=0Ameasurements are extrapolated.  This paper explores the validity of these=0Aextrapolations in the context of inter-node power variability and power=0Avariations over time within a run. We characterize power variability across=0Anodes in systems at eight supercomputer centers across the globe.  Based on=0Athese, we find that the current requirement for measurements submitted to the=0AGreen500 and others is insufficient, allowing variations of as much as 20\%=0Afrom time of measurement and a further 10-15\% from insufficient sample sizes.=0AThis paper proposes new measurement requirements to ensure consistent accuracy=0Afor power and energy measurements of supercomputers, some of which have=0Abeen accepted for use by the Green500 and Top500.
SUMMARY:Node Variability in Large-Scale Power Measurement: Perspectives from the Green500, Top500 and EEHPCWG
PRIORITY:3
END:VEVENT
END:VCALENDAR