st: Re: MC/bootstrap

[baum <baum@bc.edu>]

--On Tuesday, October 22, 2002 2:33 -0400 Stas wrote:

>     stas> A clear example of where it would be beneficial is the Monte
> Carlo /     stas> bootstrap situations. As long as both are implemented
> as ado files rather     stas> than internal commands, there would be no
> benefits from trying to run it     stas> in a multiprocessor environment,
> I am afraid :(.
I don't see the need for multithreading, Beowulf, etc. if one has a 
gazillion MCs or BS estimates to compute. I have three Unix boxes; when I 
need to do a whole lotta simulations, I just have box 1 run the simuls for 
sample sizes n1-n2, box 2 those for n3-n4, box 3 for n5-n6 and n7-n8 (since 
one of them is a multiprocessor Sun, I can run more than one Stata process 
running on it without slowing down the other). When all those jobs are 
done, I have a bunch of .dta files which Stata will be happy to append into 
the final set of MC results.

So if you have a cluster of machines, why not just split up the job? It 
would seem to me that simul and bstrap are eminently parallelizable, and 
that much simpler solutions than a parallel computing cluster are needed to 
generate the results of a simulation experiment. There are tasks which 
really require parallel computing, but it doesn't seem to me that 
independent Monte Carlo simulations are among them.

Kit

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/