Author: Robert Hyatt
Date: 13:39:24 06/13/02
Go up one level in this thread
On June 13, 2002 at 15:10:38, Bas Hamstra wrote: >On June 13, 2002 at 11:09:09, Robert Hyatt wrote: > >>On June 12, 2002 at 14:35:06, Bas Hamstra wrote: >> >>>On June 12, 2002 at 11:33:19, Robert Hyatt wrote: >>> >>>>On June 12, 2002 at 10:59:16, Bas Hamstra wrote: >>>> >>>>>Currently I use Nalimov tablebases in uncompressed format. Now I have seen you >>>>>can compress into emd format. Is it possible to use them directly in that >>>>>format? If so, how? >>>>> >>>>>Best regards, >>>>>Bas. >>>> >>>> >>>>The code Eugene supplies (on my ftp site) does this. In fact, it is one of >>>>two major innovations Eugene produced (on-the-fly decompression that is very >>>>efficient along with the very efficient indexing scheme that reduces the file >>>>sizes significantly prior to compression)... >>>> >>>>His code will recognize compressed or uncompressed tables and use either. >>> >>>Bob, 2 questions. How bad is the perfomance loss compared to uncompressed? And >>>can the 4 man tables be loaded in RAM in compressed form? >>> >>>Thanks, >>>Bas. >> >> >>Unless you have very fast SCSI disks, using compressed tables is _faster_ than >>using uncompressed tables. It reduces the total disk I/O demand since reading >>a block of compressed data is cheaper than reading a block that is not >>compressed. >> >>When I tested this way way back, using 10K rpm 160mb/sec scsi drives, not >>compressing was between 5-10% faster. But for slower SCSI drives and all IDE >>drives at the time, compressed was significantly _faster_... > >Very interesting, found "tbdecode.h" at your ftp. Last question: wouldn't the 4 >man tb be even a lot faster if loaded in compressed form in RAM? If so, why >isn't that possible (as I am told)? > >Best regards, >Bas. Under linux it happens automatically. the 3-4 piece files are so small they end up in memory buffers anyway. I suspect windows will do the same but I don't run it enough to be sure...
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.