BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20101117T220000Z
DTEND:20101117T223000Z
LOCATION:393
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: As the size of computational clusters grows, one can expect that=0AI/O will consume an increasing portion of wall-clock time=0Aas the problem and node sizes are scaled up,=0Aunless parallel I/O is introduced.=0AUnfortunately, using parallel I/O is non-trivial, so few applications=0Adeveloped by individual researchers enjoy its benefits.=0A=0AIn this paper, we describe our novel method for analyzing I/O and=0Acommunication=0Aoperations at run-time. When nodes perform I/O or communication operations,=0Aour technique protects the memory associated with the requests=0Afrom the application.=0ASubsequent operations are analyzed for overlap between=0Acommunication and I/O operations. When found, the I/O operation is=0Aautomatically transformed, by our injected library,=0Afrom an individual operation to a collective and shared MPI I/O operation.=0A=0AThis allows users to benefit from parallel file systems without=0Aredesigning or recompiling their=0Aapplications, and we demonstrate speedup for common usage patterns.
SUMMARY:Automatic Run-time Parallelization and Transformation of I/O
PRIORITY:3
END:VEVENT
END:VCALENDAR