Shared-Memory Communication in ZPL
by
Steven Smith
Abstract:
In order to keep up with Moore's Law, manufacturers of computer processors
have had to come up with new ideas for increasing the performance of their
processors. One of the more recent trends has been to incorporate
multiple processing cores within each physical processor, in effect
turning our familiar computers with only one physical processor into
multiprocessor machines. To take advantage of these new, "multiprocessor"
computers, software must be written to utilize all available processors;
programs that do not take advantage of the availability of more than one
processor will not be able to make use of the additional performance
offered by this new hardware. Work is currently in progress on a project
to create a communication layer for ZPL, an array-based programming
language, which takes advantage of multiple processors on a single
computer using shared memory. This has the advantage of avoiding the
overhead of communication schemes like MPI, which are designed to handle
communication between computers, while realizing a speed boost by running
on multiple processors simultaneously. I will describe how the standard
ZPL model had to be changed for multi-core processing, and report on
experimental progress.
Advised by Larry Snyder
CSE 403
Wednesday
March 28, 2007
4:30 - 5:20 pm