BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20131121T223000Z
DTEND:20131122T000000Z
LOCATION:601/603
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Synchronization of parallel applications like large graphs is very complex. Two major synchronization policies are used in most the architectures are- coarse grain which sequentializes program execution and fine grain which is notoriously complex. Transactional memory (TM) attempts to mitigate complexity in programming and also, provides concurrency. Parallel graph algorithms are mostly based on irregular data structures. In a large graph it is very unlikely that multiple threads will try to access the same vertex at the same time. TM benefits when there is low number of conflicts among the transactions. And, if enough number of transactions exists, they can run concurrently too. Some research results with graph benchmarks like SSCA-2 using Software based TM shows some improvements. Currently, we are working with graph benchmarks to use hardware TM support implemented in Intel Haswell architecture and believe that HTM will perform better than STM used in previous research.
SUMMARY:Transactional Memory as Synchronization Solution for Parallel Graph Exploration
PRIORITY:3
END:VEVENT
END:VCALENDAR