Author: Bruce Moreland
Date: 08:34:10 04/08/99
Go up one level in this thread
On April 08, 1999 at 10:21:21, Dave Gomboc wrote: >If you have only a low number of processors (<=4), why lock at all? Have you >tried doing _nothing_ for hash table synchronization? If the error rate isn't >too high, maybe you can get away with it. High performance increase, suffering >the chance that occasionally, you will blunder fantastically. Maybe play a >series of games and see if a blunder actually ever backs things up to the root. >(Play such a series *before* you go to the WCCC. :-) Now we are getting into the substance of this. Don and Bob were talking about hash table locking, and Don mentioned that he doesn't do it, and never notice a problem. I decided not to do it either. I can turn it on, so when I've had a bad bug I turn it on and see if it repros. So far they all have, so I know that it's not hash table locking that's killed me, yet. The hash table can always kill you, you can get a duplicate key and a wildly wrong score, so this is nothing new, it's just a little riskier than usual. There are some schemes that only lock while messing with the hash table, and there are others that need to lock when you split. I'm trying the latter variety. I do need to do locks. I haven't counted them, but I'm doing probably less than a few thousand per second, mostly uncontended. bruce
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.