Author: Ed Schröder
Date: 06:59:12 08/24/01
Go up one level in this thread
On August 23, 2001 at 23:59:05, Robert Hyatt wrote: >On August 23, 2001 at 10:41:40, Eduard Nemeth wrote: > >>On August 23, 2001 at 10:33:58, Amir Ban wrote: >> >>>On August 23, 2001 at 07:34:58, Eduard Nemeth wrote: >>> >>>>On August 23, 2001 at 07:20:04, Graham Laight wrote: >>>> >>>>I agree. But, PC from Junior was cruched, so he play on an another PC! >>> >>>Junior did not switch computers. The problem was overheating, and it was solved >>>by putting a ventillator next to the machine (thanks Jaap for suggesting the >>>solution). >>> >>>On the 21st move, after more than 20 minutes of thinking without getting a move, >>>a check showed the CPU's idle. We took an official timeout, rebooted and >>>restarted. After a few minutes this happened again. We then took a second >>>timeout, put the ventillator, and everything was fine after that. Junior lost >>>about half an hour on the clock overall. > >Did the ICCA change the way the time-out rule works? IE in the events I >have been at, it was necessary for the computer to _crash_ before it could >be re-started. I lost at least one game in the exact circumstance you >described, which sounds more like a program bug than a system problem. And >when Cray Blitz "hung" the question asked was "did the cray go down, or did >the program just deadlock?" When we showed that the cray was still up (ie >like your checking the status to see that both cpus were idle) we were told >to "sit and wait" since the machine was alive and well. We did and lost on >time. > >The thinking back then was that program bugs were just that, program bugs. And >if a program deadlocked itself, that was a problem, just as if the program went >into some sort of tight loop due to a subscript overflow or whatever. The >problem with allowing re-starts is that it is _impossible_ to restart the engine >in the same identical state it was in when it hung... > >IE the program could become convinced it had a lot of extra time due to a bug >in the time-control algorithm. It could go into a deep think, and run out the >real clock and lose, or it could be re-started which would cause the clock to >be restored to the right value as well. We always played as though the operator >was a "passive I/O device" that was not allowed to make _any_ decisions at all >such as declaring "something is broken, I have to reboot." > >I don't claim any of this is wrong. Just different than the rules used to >be applied at the ACM and WCCC events... In the Junior case the cpu usuage display said 0% which was considered as a crash. That seems ok with me. Ed
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.