[rc5] Distributed Chess camps
gindrup at okway.okstate.edu
Wed Oct 29 17:24:56 EST 1997
<*sputter all over keyboard*>
Er. It takes longer to send the data than to do the evaluation you
suggest. It would be faster for a non-distributed computer to do
this. Single plys are trivial to compute and the amount of positional
information produced thereby is also trivial. To make meaningful
position evaluations requires deep search trees. To make correct
position evaluations requires complete search trees.
This is equivalent to reverse chess: Start from every checkmate or
stalemate and produce all possible previous (legal) states. Recurse.
You do not want to calculate a single ply and then return.
These would both require 40^30 return values (as a wild estimate) or
about RC5-160's worth of network bandwidth. There does not exist a
communications infrastructure that can support 1 ply methods.
-- Eric Gindrup ! gindrup at Okway.okstate.edu
______________________________ Reply Separator _________________________________
Subject: Re: [rc5] Distributed Chess camps
Author: <rc5 at llamas.net > at SMTP
Date: 1997/10/29 13:44
Skip Huffman wrote:
> On Wed, 29 Oct 1997 10:21:41 -0500, Christopher Hodson (Consultant)
> >2. treat chess as a computer problem to be solved.
> >I find myself in camp #2, but either way, we should begin with WHAT
> >we are going to do before we try to get into implementation details
> >such as how many bits to use.
> Not exactly. Some of us want to treat chess as a *distributed*
> computer problem to be solved.
> Heck my old Atari video game from the 1970's played chess. I want to
> see how a quite complex but well understood problem can be analyzed in
> a distributed environment.
> By analyzing the issue of chess, we will be more able to understand and
> apply distributed computing to other issues. Perhaps El Ni?o.
I believe we are saying the same thing.
I propose the following 2-part algorithm for solving this chess fad:
1. Send a board description to a requesting computer which
will return all 1-ply possibilities.
2. Along each step of this tree, rate each position relative
to the previous step in the tree.
I await comments
Christopher M. Hodson
System Admin (Consultant)
Hewlett Packard Engineering Services Group
cmh at fpk.hp.com
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.
More information about the rc5