Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: New crap statement ?

Author: Robert Hyatt

Date: 07:32:09 09/25/01

Go up one level in this thread


On September 25, 2001 at 09:37:08, Dieter Buerssner wrote:

>Another idea. Often, when you fail high, actually more than one will fail high.
>An unrealistic example. Assume at a certain depth, in nodes that will fail high,
>they will either use about 10000 nodes or 1 node to fail high (because of a hash
>hit). If you search one after the other, you will randomly pick one or the
>other. On average 5000 nodes. If you search the first 2 moves in parallel, you
>will much more often get a very fast cutoff.
>



That "much more often" is wrong.  Much more often you simply search the first
move and get a cutoff, while you are searching the second.  That is called
"search overhead" as your parallel search just searched something that the
serial search avoided.  That _kills_ performance, and is the very reason no one
wants to search in parallel at any node we suspect has even a chance of failing
high.  Hence young-brothers-wait and similar algorithms...



>Have I invented ETC now? :-)
>
>Regards,
>Dieter



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.