BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20101116T231500Z
DTEND:20101117T010000Z
LOCATION:Main Lobby
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: We present a modeling framework to accurately predict time to run dense linear=0Aalgebra codes with emphasis on HPL. We report the framework's accuracy in a=0Anumber of varied computational environments such as shared memory multicore=0Asystems, clusters, and large supercomputing installations with tens of=0Athousands of cores. We also test the accuracy for various algorithms, each of=0Awhich having a different scaling properties and tolerance to=0Alow-bandwidth/high-latency interconnects.  The predictive accuracy is very good=0Aand on the order of measurement accuracy which makes the method suitable for=0Aboth dedicated and non-dedicated environments.  We also present a practical=0Aapplication of our model to reduce the time required to tune and optimize large=0Aparallel runs whose time is dominated by linear algebra computations. We show=0Apractical examples of how to apply the methodology to avoid common pitfalls and=0Areduce the influence of measurement errors and the inherent performance=0Avariability.
SUMMARY:Reducing Execution Time of High Performance LINPACK (HPL) Benchmark
PRIORITY:3
END:VEVENT
END:VCALENDAR