Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Distributed perft, current standings and trends

Author: Albert Bertilsson

Date: 23:45:14 12/12/03

Go up one level in this thread


On December 12, 2003 at 20:21:09, Sune Fischer wrote:

>On December 12, 2003 at 17:04:36, Albert Bertilsson wrote:
>
>Hi Albert
>
>I have just a few questions I've been wondering about.
>
>Since we are now using a cache, is there a time-out on each work unit before you
>submit the same WU to another machine?
>If not, what do you do if one of the machines doesn't empty its cache?
>
A very simple algo is used:
While(true)
  Try to fill up input buffer.
  If a WU exist, calculate it Else wait
  Try to send all results

So if the server lost connection sooner or later the results will be sent, the
worst thing that can happen is for the server to be offline for multiple hours
causing the clients to be idle because they have comleted all the problems in
the buffer. When the server goes online again the buffers will be
filled/emptied.

>Seems we are computing perft(3) WUs, does this mean some of them are
>transpositions?
>For example perft(8) on e4 c5 c4 is the same as perft(8) on c4 c5 e4.
>
>For perft(11) I can understand if you didn't want to optimize that, but for
>perft(12) I think it might almost cut the problem in half.
>
I did optimize it, this is called a double or a triple, for those joining early
it happened quite often that when submitting a problem their stats jumped up
two, three (or even) for problems, because the database does (pseudo code):
UPDATE Result WHERE Problem = p;
Meaning any position that matches your problem will get the result set. Dubbles
and tripples get rarer for every problem submitted, but I'm sure you'll get some
of them, they are easy to spot in the database.

>-S.



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.