The million dollar question: How well does Mogo scale on this number
of processors? Can you give us at least some kind of generalization?
unfortunately, using more than 10 nodes is probably not very very useful
in 9x9, for the moment - but we have not tested that sufficiently,
and we have not sufficiently tuned the parameters. A cluster is usefull
for tuning a sequential algorithm, but we need a cluster of clusters to
tune a parallel algorithm. Clearly, 1000 nodes for launching 40 mogo of 25
machines would be very helpful :-)
In 19x19, we can use much more - but humans are really too strong. Winning
97% against mogo is not sufficient for winning against humans who beat
mogo with probability 80%, and I'm also not sure that winning 97% against
the old mogo is sufficient for winning against CrazyStone :-)
One hope is that thanks
to parallelization, heavy playouts come for free - this is
clear in the multi-core parallelization, I guess that to some extent
multi-nodes parallelization has a similar effect for different reasons.
So, I believe in heavy playouts, whenever on sequential codes it might
be a bad idea :-)
Olivier
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
