Author: Shaun Brewer
Date: 09:28:09 10/04/02
Go up one level in this thread
On October 04, 2002 at 10:57:08, Robert Hyatt wrote: >On October 04, 2002 at 05:38:29, Shaun Brewer wrote: > >>I am trying to determine TB cache guidance for settings under Windows. This >>started because when I began using the 5 piece TB's on my laptop the disk could >>not hack it. I increased the cache setting from 8mb to 128mb as I had the ram >>available in the hope of improving things. Things appeared to get a lot worse - >>hence my original experiment. >> > > >The most important detail is that it is difficult to determine whether your disk >is >active because of excessive tablebase I/O, or paging I/O. There is _always_ a >danger of making things too large so that the system ends up paging, which is >even more harmful to speed than I/O to the various tables... > >The first thing you have to be certain about is that you are not paging in any >of >the tests... > >With larger egtb cache, crafty (on my linux box) always speeds up, so long as I >don't go "too far"... > > Thanks for your reply - unfortunately I am not paging I was very careful to ensure I stayed within real memory even when I allocated 128mb to egtb cache. Now I want a way to test the different settings under windows but I can't get the same results even when I don't change anything. So I don't know for certain if under win2k my set up etc 8mb is better than 128mb. I don't know how Linux uses spare ram for file cache but under windows I see 170mb+ allocated when using egtb cache at 8m. My hunch was that by allocating more memory to the egtb cache TB data was in effect duplicated in disk cache and egtb cache hence reducing available physical memory. Any ideas or thoughts gratefully received. Thanks again Shaun > > > > > >>However results are not consistent if I repeat the experiment. I decided to >>re-boot between to get consistent results and tested first with the same cache >>settings (Turned on the machine ran the test restarted and re-ran). However even >>that does not help see logs in seperate posts. >> >>Has anybody already done this experiment under windows or can they advise on a >>method that will provide consistent results. Should I search to a depth rather >>than for a specific time? would I be better using a different engine for the TB >>test (I have Tiger 15 as well?) >> >>Thanks >> >>Shaun
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.