Author: Robert Hyatt
Date: 07:46:38 04/06/01
Go up one level in this thread
On April 05, 2001 at 11:45:07, Ernst Walet wrote: >On April 05, 2001 at 02:17:25, Jouni Uski wrote: > >>I use normally 16MB tablebase cache definition in Fritz6 interface. I have got >>impression, that with 3+4+some5 piece tables it's good value. Yesterday I >>changed size to 0 (=no cache) and compared some endposition with a lot TB >>access. To my surprise Fritz now solved all positions faster than with previous >>value! What's this meaning? Does this mean, that Windows98 does the catching >>better than Fritz GUI? May be it's unnecessary to define any cache for Fritz. >> >>Jouni > >Using less memory for the tablebase cache leaves more memory for the disk cache >and indeed, that seems to speed the tablebase access more than a large tablebase >cache. Maybe that is because in the disk cache the data is still compressed, >while in the tablebase cache the data is uncompressed? I'm not sure but this >can be a plausible explanation. > >Ernst. I really don't believe this is possible. For several reasons: 1. Eugene's code is reached very quickly. and it is called _every_ time a hash probe is done. If it fails, _then_ we end up doing a system call, which is _very_ expensive compared to a "eugene call". 2. if you assume eugene's code keeps decompressed blocks in cache, then for every system call above, we get a compressed block and have to decompress it again. That is also non-trivial. If you reduce his cache size and things run faster although you didn't change something else, then your O/S is interfering somehow, either with paging, with poor memory layout so that cache gets aliased too badly, or something. I have run so many tests under Linux and I _never_ see better/faster results by reducing cache.
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.