Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: tablebase caching, timing, hardware rental?

Author: Robert Hyatt

Date: 09:06:33 04/07/02

Go up one level in this thread


On April 07, 2002 at 03:45:37, martin fierz wrote:

>aloha!
>
>i have a whole bunch of questions tonight. the reason is that i will participate
>in a computer checkers tournament in august in las vegas, and would like to
>improve my chances a bit :-)
>the questions are the same as in computer chess, so i hope you can help me :-)
>
>1) TB caching/access
>i just had my computer calculate the 8-piece db. in the meantime, it has also
>compressed it to about 4.5GB (20% less than the chinook db), and i'm working on
>the access code. at the moment, i use a circular buffer of database blocks, and
>run a pointer through the buffer and overwrite whatever is there if i find i
>have to access a block that is not already in memory. this works ok, but it
>seems that there are better schemes - LRU for example. could somebody explain
>how this works and how much better it would be (this is important to me! i'm
>such a lousy programmer that i'd rather not use doubly linked lists unless there

You are using FIFO it seems.  LRU is better.  A double-link would be the easiest
way to do a LRU.  When you use a block, move it to the front of the list.  When
you replace a block, replace the one on the tail of the list.  LRU is
significantly better than FIFO as a replacement policy.  It is used in every
paging facility (virtual memory) that I know of...  Eugene uses it in the probe
code used in crafty and other engines...




>really is something to gain - because i will certainly generate lots of
>bugs...)? or point me to a webpage explaining?
>also, at the moment i use VirtualAlloc to get myself a big address region which
>i use for my buffer. under windows, there are these memory-mapped files. would i
>do better to try to use memory-mapped files instead of virtualalloc? and if yes,
>my db consists of 1000s of files - can i map any number of files, or is there a
>limit?
>

No idea for windows...


>2)timing
>i never had to do anything about the timing of my program before, except for
>fixed time per move. now, we will probably play game in X. is there some kind of
>standard way to allot time per move in this mode of play? a fixed percentage of
>the remaining time maybe?


That is what I do for game in X situations.  I think I use X/25 where X is
the time left...


>also on this issue: very often, there is only one plausible move in a position
>in checkers. i guess in chess this would be similar to a position where a piece
>has been taken and only one move takes the material back. i would like to make
>such "obvious" moves faster than "normal" moves. has anybody ever experimented
>with such a thing (e.g. in the root alphabeta call, do a normal search, and then
>do a minimal window search for moves 2-n with (bestvalue-X,bestvalue-X+1)as
>window and see if it fails low for all moves, for an X of say 2 pawns? and if
>yes, stop the search for example after half or quarter the nominal search time?
>


First, define what you are calling an obvious move.  And give it some
thought.  You don't want to play into a trap, because you did a shallow
search and got tricked...

Then I suggest you use maybe 1/3 or 1/4 of the normal target time.  This
will let you move quickly while still giving you a chance to recognize trouble.






>3) hardware rental
>the checkers tournament will take place in las vegas in early august. i have a
>XP1600+ with 1GB ram here in honolulu, but getting that tower to vegas with a
>monitor is a hassle. i searched for computer rentals on the net, but most have
>rather outdated hardware. does anyone know if there is a possibility to rent a
>really good system (e.g. a bit faster than mine, as much ram as possible)in las
>vegas for a week?
>
>cheers
>  martin

One option is to take your tower, and rent a monitor...




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.