[Fwd: Re: [R] Multithreading]

A.J. Rossini rossini at blindglobe.net
Tue Apr 16 23:21:09 CEST 2002

>>>>> "timothy" == Timothy H Keitt <tklistaddr at keittlab.bio.sunysb.edu> writes:

    timothy> On Tue, 2002-04-16 at 11:22, ripley at stats.ox.ac.uk wrote:
    >> Careful!  They are running (arbitrary) R code, and Luke's caveats apply.
    >> Internally lapply does an eval(), and we can't have more than one of
    >> those running at once without all the perils of user R-level threads,
    >> as I understand it.

    timothy> Ah, yes, I see the problem there. Sounds like the rpvm version might be
    timothy> interesting though. In practice, most of the problems I work on are of
    timothy> the 'embarrassingly parallel' variety -- running the same code with
    timothy> different parameters -- and wouldn't really benefit much from low-level
    timothy> parallelism. For those, I've found that having a process query a
    timothy> database for parameter sets, compute a result and then write the result
    timothy> back to the database before proceeding to the next parameter set works
    timothy> well. You get automatic load balancing and can query the database for
    timothy> intermediate results. Maybe I'll package that up for general consumption
    timothy> at some point.

While RPVM is annoying to setup, I should mention that features that
we are working for "Release Soon"(tm), are
1. use of the SPRNG library, as a separate package (this is a parallel
   RNG, which gives reasonable results and is designed for parallel
   situations).  This might be useful in other contexts as well
2. simple examples for bootstrap.

I should note that someone (to remain nameless) has offered to provide
a means to serialize R objects sensibly.  We'll see when that happens;
it's simple provided that you program carefully.  However, I'd like a
variant which is simple even when you are a programming klutz.

I'd also like to construct simple examples for simple global
optimization as well as MCMC, but I don't see doing these personally
in the near-ish future.

In addition, note that PVM jobs migrate on MOSIX/OpenMOSIX.  I've
preliminary evidence that this is true for RPVM jobs, as well (since
Python is problematic under MOSIX, I'm not going to claim more than
preliminary evidence). 


A.J. Rossini				Rsrch. Asst. Prof. of Biostatistics
U. of Washington Biostatistics		rossini at u.washington.edu	
FHCRC/SCHARP/HIV Vaccine Trials Net	rossini at scharp.org
-------------- http://software.biostat.washington.edu/ ----------------
FHCRC: M-W: 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
(my friday location is usually completely unpredictable.)

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

More information about the R-help mailing list