BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20111115T234500Z
DTEND:20111116T000000Z
LOCATION:TCC LL1
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: In the Exascale era, the evolution of supercomputing will bring massive machines comprising millions of cores and a similar number of components. In those complex machines adaptivity will be fundamental. One dimension of adaptivity, fault tolerance, is considered one of the main challenges for systems of that size. Some estimates project an Exascale machine will fail several times per hour. To deal with a high frequency of failures, it is necessary to go beyond traditional checkpoint/restart approaches, where all processing elements are rolled back to their latest checkpoint even after the crash of only one of them. Message logging is a promising alternative which, nevertheless, faces various challenges. This work addresses the major drawbacks of message logging and provides a set of techniques that will make it an effective solution to provide resilience for Exascale machines.
SUMMARY:Scalable Message Logging for HPC Applications
PRIORITY:3
END:VEVENT
END:VCALENDAR