BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20131120T001500Z
DTEND:20131120T020000Z
LOCATION:Mile High Pre-Function
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Exascale computing pose a number of challenges to application=0Aperformance.  Developers need to study application behavior by=0Acollecting detailed information with the help of tracing toolsets. But=0Anot only applications are scalability challenged, current tracing=0Atoolsets also fall short of exascale requirements for low background=0Aoverheads since trace collection for each execution entity is becoming=0Ainfeasible. One effective solution is to cluster processes with the=0Asame behavior into groups. Instead of collecting performance=0Ainformation from all individuals, this information can be collected=0Afrom just a set of representatives. This work proposes a fast,=0Ascalable, signature-based clustering algorithm that clusters processes=0Athat exhibit the same execution behavior. Instead of prior work for=0Astatistical clustering metrics, it produces precise results without=0Aloss of events or accuracy. The proposed algorithm combines log(P)=0Atime complexity, low overhead at the clustering level, and it splits=0Athe merge process to make tracing suitable for exascale computing.
SUMMARY:Scalable Performance Analysis of Exascale MPI Programs through Signature-Based Clustering Algorithms
PRIORITY:3
END:VEVENT
END:VCALENDAR