Re: [Bug-gnubg] multiprocessor on linux

[Jim Segrave]

On Tue 02 Dec 2003 (15:25 +0100), Achim Mueller wrote:
> Hi folks (especiallay Joern),
> 
> I just got access to a compaq dl 580 g3 with 4 xeon 2.8ghz and
> hyperthreading activated. I would like to do some work using smp.
> 
> What do I have to do for activating and using this feature? For now
> a rollout uses only one CPU.

Start coding. Come back in a month or two :-)

I actually think that building a no-interface analysis/rollout only
version of gnubg (getting all the needed settings from a master) is a
better approach than trying to make gnubg threaded. It also works both
when using a network of machines or a multi-cpu machine. Most OSes
would soon migrate separate copies to separate CPUs on an SMP machine
and there are no locking issues to resolve, only simple
synchronisation ones.

For analysis, issuing separate moves to separate clients would work
reasonably well, although some caching would be lost. For rollouts,
passing a starting position and the rollout settings to each client and
then doing a rollout locally and waiting for the clients to complete
is the best method (needs keepalives and timeouts to prevent one
client stopping from hanging the system), but this would allow for
stopping the rollout on the master and getting to a consistent state
for saving a rollout so it could be extended.

-- 
Jim Segrave           address@hidden