Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Bean Counter

Author: Dann Corbit

Date: 12:39:01 09/29/99

Go up one level in this thread


On September 29, 1999 at 15:22:02, Will Singleton wrote:
[snip]
>I don't have any experience with parallel processing, so I can't really comment
>much.  But it appears that by abandoning "dead-ends" near the root, you are
>talking about a very selective searcher, not full-width.  Or do you equate
>"dead-ends" with cutoffs?
It's a combination.
The one thread, bolted to the root does a full width search.  The other threads
are doing "optimistic hunting".  Let's say I have 30 positions forward from the
current node.  The heartbeat is set at 30 seconds.  After 30 seconds, the 30
worker threads check back in.  The 27 with the worst evals abandon their tasks
and join the effort with the 3 best -- looking at children of those nodes.

Ideally, I will have hundreds or thousands of processors.  The root node will be
a special thread and have high priority and always hold a complete processor all
to herself.  The other ones will be a cloud of bees looking for nectar.




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.