R devel - Mar 1999 - re: smp in Linux

A question to all you R-gurus:

Can R (or S-plus, for that matter) make efficient use
of multiple Intel Processors running under Linux (within 
the same PC, not over a net)? 
With the release of the new 2.2 kernel, this would seem 
a interesting and cost-efficient way of boosting the 
computational power of Intel/Linux platforms when using 
R (or S-plus).

Thanks for any wise words,

Kenneth Nordstr?m

.....................................................................

Kenneth Nordstr?m                      Kenneth.Nordstrom at oulu.fi
Professor of Statistics
Dept of Math Sciences/Statistics       Tel: +358-8-553 1829 
Univ of Oulu                                +358-8-553 1820 (secr) 
P.O. Box 3000                          Fax: +358-8-553 1848
FIN-90401 Oulu
FINLAND
 
.......................................................................
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

On Wed, 10 Mar 1999, Kenneth Nordstrom wrote:
 
> A question to all you R-gurus:
> 
> Can R (or S-plus, for that matter) make efficient use
> of multiple Intel Processors running under Linux (within 
> the same PC, not over a net)? 
> With the release of the new 2.2 kernel, this would seem 
> a interesting and cost-efficient way of boosting the 
> computational power of Intel/Linux platforms when using 
> R (or S-plus).
R is not currently multithreaded or otherwise parallelised. Threading is
in the list of things that would be interesting to do sometime in the
distant future. It will be hard.

There are possibilities for parallelisation short of making R fully
threaded.  One more-or-less straightforward one would be to replace the
matrix operations with efficient parallel code. This wouldn't help a lot
since R doesn't spend that much time doing matrix operations.  Another
possibility would be to run the graphics drivers in parallel with the main
code. This wouldn't help much, either, unless you routinely draw
high-resolution images or overloaded scatterplots.

A third possibility, which would be more difficult but might actually gain
something is to thread the "findVarInFrame" search. I did some
profiling
last year sometime and this was taking up 20% of the execution time on
the examples I looked at. Cutting this in half might actually be 
noticeable.

On the other hand, most of the problems I have where speed is an issue are
simulations or bootstrapping problems that scale almost perfectly by
running 1/n as many on each of n processors.  I have managed 190% CPU
utilisation by two copies of R on a dual Pentium Pro even under an older
Linux kernel.


Thomas Lumley
Assistant Professor, Biostatistics
University of Washington, Seattle

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

>On the other hand, most of the problems I have where speed is an issue are
>simulations or bootstrapping problems that scale almost perfectly by
>running 1/n as many on each of n processors.
This has been my experience too, and I have lots of problems like this. It would
sure be nice to have some possibility like:

parallelfor (i in 1:N, listofmachines )  x[[i]] <- simulate(model[[i]],
data[[i]])

Paul Gilbert

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

On Thu, 11 Mar 1999, Paul Gilbert wrote:
> >On the other hand, most of the problems I have where speed is an issue
are
> >simulations or bootstrapping problems that scale almost perfectly by
> >running 1/n as many on each of n processors.
> 
> This has been my experience too, and I have lots of problems like this. It
would
> sure be nice to have some possibility like:
> 
> parallelfor (i in 1:N, listofmachines )  x[[i]] <- simulate(model[[i]],
> data[[i]])
> 
This may be a case for implementing .For().  We currently say it is not
necessary and thus not implemented, which is true for systems with more
users than processors, but it would be useful for multiprocessor systems.

	-thomas


Thomas Lumley
Assistant Professor, Biostatistics
University of Washington, Seattle

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._