A question for the computer nerds on the list--can Quackle exploit a
multicore or multiprocessor environment by doing the simulations via
large-grain parallelization?  I.e. running each sim on a different
core insofar as possible?  I think that requires looking ahead at the
random number generator, which seems like a hard problem to me, but I
was wondering if anyone has a clever solution to it.  I.e. suppose I
want to run sim#1 with seed 4567654 and sim#2 with seed 87654567
because the seed at the completion of sim#1 would be 87654567, so you
need to know how many draws on the random number generator sim#1 will
do, then step forward through all of them to get the new seed.  Yes?
If there were a max # draws on the random number generator any sim
could do, you could step forward by that offset every time to get the
new seed, I think, and then the question is whether that process of
stepping forward is a greater expense in terms of processing than
justifies the speed advantage of moving even-numbered sims to the
second core.   I am from a different background in computing,
obviously, so I don't know if the question even makes sense to
developers on this list.  Jason, Olaugh, others, does this question
make sense to you?

On Sat, Aug 16, 2008 at 8:47 AM, G. Vincent Castellano
<[EMAIL PROTECTED]> wrote:
> Actually, that was pretty easy.  Good work, guys.
>

Reply via email to