Title: Multi-core processors: how to program them effectively?
Speaker: Daniele P. Scarpazza
Cell Solutions, IBM T.J. Watson Research
Abstract:
Multi-core processors are revolutionizing the world of computing: the
growths of frequency and complexity seem to have hit insurmountable
physical walls, and thus designers are instead putting more and more
cores on a single chip, in an attempt to provide ever-increasing
computing power. This focus on throughput and parallelism represents
a major shift in the design of mainstream computers. The IBM Cell/B.E.
processor is a highly representative, early example of this growing
trend, with its nine cores and its impressive arithmetic throughput of
more than 100 billion double-precision floating point operations per
second (or 200 billion single-precision ones).
Is this growth sustainable? What kind of multi-cores will we see in the future? Are traditional programming models capable of exploiting this power? Are software developers ready to "think parallel"? How will we program multi-cores in five years?
To answer these questions, Daniele will discuss in detail some of the lessons learned on the IBM Cell/B.E. processor from exploring algorithms such as graph exploration, string and regular expression matching, information indexing, and more.
Daniele Scarpazza's Short Bio:
Daniele is a Post-Doctoral Research Fellow at the Cell Solutions Department of the IBM Thomas J. Watson Research Laboratory in Yorktown Heights, NY. He works on parallel algorithms and programming abstractions for multi-core processors, with an emphasis on pattern matching, information retrieval, and efficient management of large data structures.