ZPL Programming Language

Greg Lindahl lindahl at conservativecomputer.com
Wed Apr 25 14:13:12 PDT 2001


On Wed, Apr 25, 2001 at 01:30:40PM -0700, JParker at coinstar.com wrote:

> While surfing the web, I came across a programming language developed at 
> the University of Washington for parallel applications.
> 
> http://www.cs.washington.edu/research/projects/zpl/

There are a zillion similar projects. While I like some of ZPL's
religion, it suffers from some of the usual flaws. First, it depends a
lot on array expressions, but many algorithms aren't neatly expressed
in array expressions. Second, it has many of the same compilation
problems that HPF has, although it does dodge some. As a result,
speedup is limited.

In their performance papers, they seem to spend more time saying "We
can do better than HPF" than "We can get a speedup of 100x on real
algorithms." Consistant with that, the performance paper I looked at
doesn't even make it easy to read the speedup off the chart.

So: If you have a problem that's neatly expressed by array syntax,
maybe you'll get some speedup. But "real world" tool (like SMS, which
I flog on a regular basis: http://www-ad.fsl.noaa.gov/ac/sms.html) can
give far better speedup on clusters, if your problem fits the
tool... and there seem to be real problems which fit such tools, more
often than problems which do well in ZPL.

-- g




More information about the Beowulf mailing list