BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20141118T231500Z
DTEND:20141119T010000Z
LOCATION:New Orleans Theater Lobby
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Communication is a scaling bottleneck for many parallel applications executed=0Aon large machines. Intelligent task mapping may alleviate the negative impact=0Aof communication, but simple metrics used to find such mappings may not be good predictors of their performance. We evaluate supervised machine learning=0Amethods as tools for prediction of communication time of large parallel=0Aapplications. Through these methods, we correlate communication time for=0Adifferent task mappings to the corresponding network hardware counters=0Aaccessible on the IBM Blue Gene/Q. The results from these machine=0Alearning regression algorithms are used to provide insight into the relative importance of different hardware counters and metrics for predicting application=0Aperformance. These results are explored graphically in the poster in the=0Acontext of two production applications, MILC and pF3D.
SUMMARY:Supervised Learning for Parallel Application Performance Prediction
PRIORITY:3
END:VEVENT
END:VCALENDAR