BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20111116T000000Z
DTEND:20111116T001500Z
LOCATION:TCC LL1
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: This research addresses three main sources of challenge in debugging parallel applications. First is the cognitive challenge where programmers cannot integrate the data effectively into a mental model of correct execution, due to the scale of the state of a typical scientific application. Second, traditional control-centric debugging approach becomes unwieldy when thousands of compute threads involved. Finally, typical parallel debuggers generally don't utilize the underlying parallel platform to improve their own performance.=0AThis research argues a more data-centric debugging approach, in which adhoc debug-time assertions can be used to test the state of a large distributed array, and can be refined iteratively to locate program’s defects. Furthermore, this research proposes a scheme that executes the assertion in parallel, making assertions over large data structures feasible.=0AThe implementation of different types of debug-time assertions, and various case studies, along with thorough performance evaluation results have been presented in several publications.
SUMMARY:Data-centric Parallel Debugging Technique for Petascale Computers
PRIORITY:3
END:VEVENT
END:VCALENDAR