Author: Kolss
Date: 06:25:55 12/14/04
Go up one level in this thread
On December 14, 2004 at 08:30:13, Francesco Di Tolla wrote: >>I don't think that is a good idea: when an engine is pondering, but does not get >>any CPU time, it will still think that it has been searching for the whole time. >>If e.g. after a ponder hit, an engine substracts the time used during the >>opponent's turn from its move time, it may move instantaneously (or very fast), >>assuming that it has been searching for a decent time interval, although >>effectively it has not. That should obviously severely impair the performance... > >this would be a cheap way to decide for a move. One should rather check the >depth. > >How can you prevent silly horizont effects if you simply check the wall clock? Well, it is just that I would guess that most programs handle it this way! Under normal conditions, this is a very safe strategy. E.g., if the program makes a move, starts pondering, and after (say) 12 minutes (as might happen in a tournament game) the opponent makes the expected move, it is mostly (i.e. if no fail-low, etc.) reasonable to move immediately, since the program has had plenty of time to think. This is correct as long as it actually got the CPU, as would be the case if there were only one program running on the computer (CPU). But in the case of idling the process, the program still makes the (normally) reasonable assumption that it has been searching for 12 minutes (at full power!), while in fact it has not done anything. I wonder how many programmers have thought of this situation when implementing time management with relation to pondering. I certainly have not... Best regards - Munjong (cheapo programmer...?! :-).
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.