[parsec-users] A problem about Parsec Benchmark

Nikolay Kurtov nkurtov at gmail.com
Mon Sep 15 12:23:54 EDT 2008


Hi Shinra,

The paper http://parsec.cs.princeton.edu/doc/parsec-report.pdf perfectly
describes purpose of each pipeline stage.

Using pipeline model computations can be made more parallel. Making the
algorithm only data parallel will bring additional constraints, that will
cause unnecessary serialization of independent "high-level operations".
The better you express the parallelism the more scalable will be the
application.

The main headache is the last stage that serially assembles the output :)

Good luck!

--
Nikolay

mailto: nkurtov at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20080915/0c14d3ff/attachment.html>


More information about the parsec-users mailing list