Re: [Bug-gnubg] gnubg

[Joern Thyssen]

On Sat, Jun 15, 2002 at 09:31:06PM -0400, Julian Haley wrote
> I've recently started using gnubg and have found the program to be of
> great quality (not to mention its immense value to the bg community!)
> 
> 
> My only suggestion would be to allow the program to take advantage of
> multiple processors. This is a feature I have not seen on any other bg
> analysis tool and could be very helpful in speeding up the analysis
> for some people.

Support for multiprocessors is, in fact, noted in the TODO list.

There are quite many very different systems to support:

- SMP (e.g., dual-processor linux/windos systems, SGI, Sun, and IBM 
  RS6K)
- distributed systems (e.g., beowulf clusters)
- distributed SMP systems
- probably more

The first problem will most likely be that each multiprocessor system
requires platform specific code. It'll probably be difficult to make
some platform independent code. Of course, there exist attempts to
standardise SMP with OpenMP and distributed systems with MPI. There is
also pthreads and other alternatives.

The next problem is access to these kind of systems -- I'm not sure how
many of the developers who have access to multi-processor systems?!

One may also parallelise at a high level (e.g., rollouts and high-ply
evaluations) or at a low level (neural net evaluations). It's will
probably most efficient and easiest to program parallelism at a high
level.

I think it would be very trivial to parallelise rollouts with ptreads,
but we would have to go through the source and check if it is
thread-safe everywhere.

Joern

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., 9000 Aalborg
+45 9813 2791 (private) / +45 2077 2689 (mobile) / +45 9633 7010 (work)