[R] help with random numbers and Rmpi

[A.J. Rossini]

Faheem Mitha <faheem at email.unc.edu> writes:

> On Sun, 14 Dec 2003, A.J. Rossini wrote:
>
>>
>> Did you make sure that you loaded the library that contains the C
>> code on the remote hosts?
>
> No, I only loaded the library on the local node (master).
>
> However, I'm not sure how I should do so. If I simply start up R on the
> remote hosts and load up the library, I presume this would not be any use,
> since the master spawns its own R slaves?

No, it would be of use.  The approach is to  run multiple independent
processes -- this is the general "message-passing" paradigm, and holds
true regardless of the particular API being used.  So each R process
has to be appropriately initialized.   SNOW is taking care of the RNG,
but you've got to do everything else.

> So, the question would be how to get the R slaves to load up the library
> and I don't see any obvious way of doing this (using snow functions).
>
> I must be missing something. Do I need to work at a lower level?

Yes, and no.

Re-read the CPH-statcomp lab, and look at the bootstrap example, which
solves the same problem.

Look carefully -- it has to initialize the library on each node.
If you are just loading the library manually, just do it on each node;
if you are using libraries, just do that.

best,
-tony

-- 
rossini at u.washington.edu            http://www.analytics.washington.edu/ 
Biomedical and Health Informatics   University of Washington
Biostatistics, SCHARP/HVTN          Fred Hutchinson Cancer Research Center
UW (Tu/Th/F): 206-616-7630 FAX=206-543-3461 | Voicemail is unreliable
FHCRC  (M/W): 206-667-7025 FAX=206-667-4812 | use Email

CONFIDENTIALITY NOTICE: This e-mail message and any attachme...{{dropped}}