Re: [gnugo-devel] multi-threaded GNU go

[Terry McIntyre]

Christian,

You might consider working with Monte Carlo Go engines; have you looked at
the Computer Go discussion list and the Computer Go Server?

http://computer-go.org/pipermail/computer-go/
 - Computer Go discussion archives

http://cgos.boardspace.net/ - Computer Go Server, an automated tournament amongst many computer Go programs; lately, various MC programs, particularly Mogo, have been doing very well. I suspect that a good 64 processor implementation would take the top position.

David Doshay has done some work with a sort of multi-processor extension built on GnuGo.
http://sluggo.dforge.cse.ucsc.edu/harryMS.pdf and other papers

I suspect that MC/UCT programs may be a better fit for a multi-processor architecture; as some have already observed, there's a lot of state to synchronize with gnugo. MC/UCT has the virtue of having a simpler design, which may be easier to parallelize.

Best of luck! The future is definitely multi-processor; just about everyone is buying dual-core nowadays, quad-core will be widely available by the end of the year, and in a few years, dozens of processors are likely to be the norm.

An interesting research field at the moment is to incorporate some intelligence into the "random" playouts of monte carlo simulations. The gnugo engine might prove to be a fruitful source of ideas in that regard.

On 2/20/07, Christian Bienia <address@hidden> wrote:

Hi,

> > I was the one trying to implement multi-threaded branch [1].  With all due
> > respect, I doubt it is doable without good knowledge how the engine is
> > implemented internally.  E.g . you need to synchronize or split to one per
> > thread several caches that exist etc.  The branch never reached working
> > state as I abandoned it, as far as I remember after unpromising timing
> > results.
> >
> > [1] cvs -d:pserver:address@hidden:/sources/gnugo -r multi-board co gnugo
>
> See the list archives around April 2005.
>
> http://lists.gnu.org/archive/html/gnugo-devel/2005-04/

That's too bad. :-( Multi-threading would be a great enhancement for GNU
go. I suppose I'll have to find something else for the students. Any
ideas for alternatives will be appreciated a lot.

Are there any plans to reconsider that decision, given the fact that the
two major chip producers will stop producing single-core processors this
year? As Darren pointed out:

> the future is multi-core not higher clock speeds. Unfortunately.

- Chris



_______________________________________________
gnugo-devel mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/gnugo-devel

-- 
Terry McIntyre
UNIX for hire
Software Development, Systems Adminstration, Securiity
address@hidden
213-291-5169