Author: Robert Hyatt
Date: 17:36:24 05/12/01
Go up one level in this thread
On May 12, 2001 at 10:44:46, Frank Phillips wrote: >On May 12, 2001 at 09:42:15, Robert Hyatt wrote: > >>On May 12, 2001 at 03:22:30, Frank Phillips wrote: >> >>>When my program is running on FICS under Linux and accessing the EGTBs >>>(performing a lot of disk activity), it seems to forget to look at the input >>>from the server for long periods of time, so that the opponent's clock goes >>>negative. When not accessing the EGTBs (or running under Windows) everything is >>>fine. >>> >>>(Frustrating, since I may have to move to Linux if MS decide next that your >>>first born is part of the licence fee for using their sotware, in interests of >>>the consumer and to protect freedom and democracy against the evil threat of >>>open source software of course.). >>> >>>Any help appreciated. >>> >>>Frank >> >>There must be something else wrong. I run linux exclusively and have _never_ >>seen anything like this, period. I find linux scheduling and interrupt handling >>to be as good as anything around. You are not useing a node counter to trigger >>your input check I hope? IE if you decide to check every 500K nodes, and you >>reach an ending where you only search 50K nodes per second, you might not be >>checking for input often enough... > >I am using a node counter to trigger input check and I do see the dramatic drops >in nps you indicate, but set the trigger value according to previous nodes per >second (unless there is little time left), so should only see the described >behaviour once. Once _could_ be enough. And if you have very slow IDE access, things could really get sluggish no matter what the O/S. > > >/* >Calculate number of nodes before a test is done for user input on | >| next itereation. SetTargetTime() sets time_check_nodes to | >| last_time_check_nodes or shorter depending on time remaining. | >| | > ----------------------------------------------------------------------- >*/ >int time=(root.stop-root.start); >if (time > 100) > last_time_check_nodes=stat.nodes/time*100/4; //4 times per second. >else > last_time_check_nodes=TIMECHECK; > > > > >When I understand threads better I will try that approach....... > >I thought Crafty used a node counter in Search(). Better take another look. > >Frank I do... but I control the NPS slowdown so that I don't see drops from 1M to 20K, ever... . I rarely drop below 500K in fact.
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.