Cheaply constructing "high-level" parallel computations Re: Just A Few Questions... (One massive Reply)
> On Thu, Mar 13, 2003 at 07:34:46PM -0500, myuu wrote:
>
> :> gridmathematica : :That would be cool, but I live in Bismarck, ND;
> :> no store is going to selling software like that -- even if it did, I
> :> wouldnt be able to afford it. And i doubt ill find it on any P2P
> :> progs, wolfram has a heck of a license test.
We've finished putting together a simple framework for parallelizing
computations in R, a GNU GPL licensed statistical language.
Currently, it supports PVM, LAM-MPI, and a socket implementation as a
backend. It might be a bit heavy mathematically to use for demos
(unless words like "bootstrap", "profile likelihood", "spatial
prediction/kriging" and "cross-validation" are in your daily vocab
:-), but we've got a tech report on using it.
http://www.bepress.com/uwbiostat/paper193/
and some details (via a noweb processed file) at:
http://www.stat.uiowa.edu/~luke/R/cluster/cluster.html
and you would need to install:
Rmpi, rpvm, snow, and rsprng (available at http://cran.r-project.org/,
or via R's "install.packages()" function, and the SPRNG numerical
libraries as well for minimal guaranteed parallel random number
generation, to make it work. There is an example and details as well
in the technical report and on Luke's WWW site.
It's sort of cool (though silly) to evaluate problems on PVM and MPI
virtual clusters at the same time (on the same set of
machines/physical nodes!).
best,
-tony
--
A.J. Rossini Rsrch. Asst. Prof. of Biostatistics
U. of Washington Biostatistics rossini@u.washington.edu
FHCRC/SCHARP/HIV Vaccine Trials Net rossini@scharp.org
-------------- http://software.biostat.washington.edu/ --------------------
FHCRC:Tu: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email
UW: Th: 206-543-1044 (fax=3286)|Change last 4 digits of phone to FAX
(CHANGE: monday/wednesday/friday locations are completely unpredictable.)
Reply to: