I'd be interested in solutions, as well -- we've used SPRNG, but
havn't really had a good sense of it's properties. A recent article
in one of the optimization/scheduling journals suggested a better
approach, but havn't had a chance to really track it down.
sirola_at_[hidden] writes:
> hello,
> this is not strictly related to LAM/MPI but to parallel programming in
> general, anyway probably many subscribers solved this problems
> already.
> I should implement a parallel montecarlo on a cluster. There will be a
> crowd of processes computing "paths" and every process will obviously
> have its random number generator, now the problem is being "sure" that
> the generated random sequences won't be correlated. I've considered 2
> RNG:
> lagged fibonaci
> mersenne twister
>
> mersenne-twister tuned for parallel montecarlo seems
> very good (basically you supply the process rank as a seed and
> everything is ok), anyway this RNG has a very slow bootstrap process
> (1 min more or less on a p3) so i'd use the lagged Fibonacci
> RNG (a MC path will use few thousands numbers and i will have 100k
> path maximum). I've read you can use a linear congruential generator
> to generate the lagged fibonacci seeds, but i'd need some further
> information: does any1 have any refererence/pointers about this,
> or have a simple piece of code to show?
> Thanks in advance for your help,
> enrico
> --
> Enrico Sirola <sirola_at_[hidden]>
> gpg public key available from www.keyserver.net, Key ID 0x377FE07F
> Key fingerprint = B446 7332 ED55 BC68 5FE8 DE0F 98DF EC86 377F E07F
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
A.J. Rossini rossini_at_[hidden] http://software.biostat.washington.edu/
Biostatistics, U Washington and Fred Hutchinson Cancer Research Center
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
CONFIDENTIALITY NOTICE: This e-mail message and any attachments may be
confidential and privileged. If you received this message in error,
please destroy it and notify the sender. Thank you.
|
