Author: Robert Hyatt
Date: 16:59:58 11/12/02
Go up one level in this thread
On November 12, 2002 at 10:56:15, Russell Reagan wrote: >On November 12, 2002 at 10:26:46, Uri Blass wrote: > >>I think that they may get 5 plies and not 2 plies and >>I also think that 2 plies can provide useful information. >>For example you may see that all the moves except 2 are losing so you can >>increase the priority of the interesting thread(not the move you expect) to 10% >>and you have 90% for the move that you expect and 10% for an interesting move to >>check. >> >>if you search 10-12 plies with 90% of the time then you may search 8-10 plies >>with 10% of the time and you can increase the 10% to more than it later based on >>information that you get in the search. > >You need to learn about how threads work. You can't set them to percentages. You >can only set them to a handful of levels. For example, you can set thread A to >run at a priority of NORMAL, and you can set threads B, C, and D to run at >BELOW_NORMAL, and thread A will ALWAYS run before B, C, or D. That means that if >thread A has something to do all of the time (and if you were pondering, it >would) threads B, C, and D would NEVER get ANY processing time, so when I said 1 >or 2 plies for the low priority threads, that was a generous estimate on my >part. In reality it probably wouldn't get any processing time at all. This is >how it works in Windows at least. I'm also not sure what the maximum number of >threads is, but there is a limit. I know that WaitForMultipleObjects will only >handle a maximum of 64 threads. Or on linux you could use nice. Where nice 20 means that thread will get about 5% of the cpu, nice 0 (the default) means it competes equally with other processes, and the others vary over that range. But I think that idea is simply bad for lots of reasons...
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.