MIT Lincoln Researchers Release pMatlab (fwd)

[Barry Smith]

MIT Lincoln Researchers Release pMatlab
HPCwire M489132
October 14, 2005

For years, users of high performance computing users have struggled
with difficult low-level message passing programming environments to
exploit parallel computers.

MIT Lincoln Laboratory researchers Jeremy Kepner, Nadya Travinin,
Albert Reuther and Hahn Kim -- under the sponsorship of the Department
of Defense's High Performance Computing Modernization Program, the
Defense Advanced Research Projects Administration and the Air Force --
have been leading an effort to bring MATLAB -- a common high-level
language used in technical computing -- into the HPC world.

The first stage of this effort was the development of the MatlabMPI
library (<http://www.ll.mit.edu/MatlabMPI>), which now has thousands
of users. More recently, this team has developed the pMatlab library
(<http://www.ll.mit.edu/pMatlab>), which brings the benefits of
Parallel Global Array Semantics (PGAS) to the Matlab
community. Currently, PGAS is mostly found in specialized HPC
languages, such as UPC, Co-Array Fortran and Titanium, as well as
numerous C++ libraries (e.g. POOMA, GA++, and ||VSIPL++).

pMatlab allows users to describe how to break up their problems on a
parallel computer using extremely compact and natural array semantics
that are intuitive to most scientists and engineers. pMatlab allows
users to write vastly more complex parallel codes with a lot less
work. For example, pMatlab implementations of the HPCChallenge
benchmark suite (http://www.HPCChallenge.org
<http://www.hpcchallenge.org/>) were between two and 60 times smaller
while delivering comparable performance. Most importantly, novice
users can convert serial Matlab programs to parallel programs in an
average of less than two hours.

According to Lincoln Laboratory's Grid Computing lead Albert Reuther,
"We have over 80 pMatlab users at Lincoln. We have never seen a
MatlabMPI user go back after they have tried pMatlab."

One example of the impact of pMatlab was a recent flight test where
pMatlab was used to quickly process research sensor data on a
28-processor blade cluster on a Boeing 707 aircraft.




-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (Cygwin)
Comment: Processed by Mailcrypt 3.5.8 <http://mailcrypt.sourceforge.net/>

iD8DBQFDT8uOmWA2y7s1YXMRAg0UAJ0Ywtvcncf84X2Gfg7hurnWTZ6BMwCgoFFf
3mwzWHvzmxmiesxUEEYaYFk=
=TlNf
-----END PGP SIGNATURE-----