Author: guy haworth
Date: 10:25:18 09/14/99
Go up one level in this thread
Robert: again, tfy reply with which I agree.
I have a copy of L.B.Stiller's thesis with the data about 6-man endgames he
tackled [# of positions, max-depth to conversion (not mate) and # of zugzwangs]
if you want me to post it here.
LBS had the following results for KQQKQQ:
5,122,186,896 positions (32 of which are mutual zugzwangs)
conversions are 44m deep or less
83% are won by the first player which sound ominous but ...
the vast majority being wins in
1m (1ply) (Bl. has a Q en prise),
2m (3p) (White checks the bK which moves to reveal an en prise bQ)or
3m (5p) (exercise for the reader!).
I think there was data about memory size available on the Connection Machine he
used; as you say, primary store was the thing that made this possible. I don't
know what store sizes are typical on uni- or multi-processor (Unix presumably)
machines.
C.Wirth & J.Nievergelt (ICCA Journal, v22.2 and a very fine piece of work and
paper) used parallel-processing (see p71) and a synergy of their ideas and
Eugene's on implementation would be interesting. They are, I see from ICCA J,
on wirthc@inf.ethz.ch and jn@inf.ethz.ch
My 10-years-ago experiences with MIMD machines and multiple disc-channels led me
to believe that hashing-randoming the keys and distributing the data to 2^n
stores on the basis of the leading n-bits made best use of access space,
'collisions' being picked up locally. This was the basis of the 'SQL JOIN unit'
(or File Correlation Unit as we called it) for the CAFS engine. Better to fry
2^n discs than 1.
So, perhaps we may look forward to some interesting progress.
Best regards to all, Guy [:]-)
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.