[9fans] parallelizing/distributing quasi-Monte Carlo

[9fans] parallelizing/distributing quasi-Monte Carlo

[Joel C. Salomon]

We've been using quasi-Monte Carlo integration in a class here at
Cooper, and the discussion turned to parallelization.  Some of my
classmates are going the MPI/cluster route, and I was thinking this
might be a good use of Plan 9's CSP style.  The general idea I had
uses one point generator process, n function evaluation processes, and
a collection process that handles estimation &c.

I think I know how to structure this on a single multiprocessor
machine, but what about a cluster?  Is there the equivalent of alts on
network connections?

If the question is unclear, I'll try again after I've gotten my
machine running the single computer form.

--Joel

Re: [9fans] parallelizing/distributing quasi-Monte Carlo

[erik quanstrom]

> I think I know how to structure this on a single multiprocessor
> machine, but what about a cluster?  Is there the equivalent of alts on
> network connections?

cec(8) muxes the keyboard and the network in a simple fashion.
/n/sources/contrib/quanstro/src/cec/mux.c.  this could be written
with Chan so that one could alt on network messages.

- erik