Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Why does tiger lose games on time?

Author: Robert Hyatt

Date: 07:04:24 12/13/99

Go up one level in this thread


On December 13, 1999 at 02:04:47, Lex Loep wrote:

>On December 12, 1999 at 21:02:18, Robert Hyatt wrote:
>
>>On December 12, 1999 at 17:43:57, Lex Loep wrote:
>>
>>>On December 12, 1999 at 15:43:54, Robert Hyatt wrote:
>>>
>>>>On December 12, 1999 at 15:27:19, Lex Loep wrote:
>>>>
>>>>>On December 12, 1999 at 10:10:58, Robert Hyatt wrote:
>>>>>
>>>>>>On December 12, 1999 at 09:33:38, blass uri wrote:
>>>>>>
>>>>>>>On December 12, 1999 at 07:27:27, Lex Loep wrote:
>>>>>>>
>>>>>>>>On December 12, 1999 at 06:00:43, blass uri wrote:
>>>>>>>>
>>>>>>>>>On December 12, 1999 at 05:31:47, Lex Loep wrote:
>>>>>>>>>
>>>>>>>>>>On December 12, 1999 at 03:23:18, blass uri wrote:
>>>>>>>>>>
>>>>>>>>>>>On December 11, 1999 at 23:38:05, Chessfun wrote:
>>>>>>>>>>>
>>>>>>>>>>>><<SNIP>>
>>>>>>>>>>>>I posted the previous score as being:
>>>>>>>>>>>>Record for shutka vs. chesspartner:
>>>>>>>>>>>>                          wins     losses     draws
>>>>>>>>>>>>                rated         60         29         0
>>>>>>>>>>>>              unrated          0          0         0
>>>>>>>>>>>>
>>>>>>>>>>>>Current is now:
>>>>>>>>>>>>Record for shutka vs. chesspartner:
>>>>>>>>>>>>                            wins     losses     draws
>>>>>>>>>>>>                rated         61         30         0
>>>>>>>>>>>>              unrated          0          0         0
>>>>>>>>>>>
>>>>>>>>>>>I am surprised to see this result mainly because of the fact that they are no
>>>>>>>>>>>draws.
>>>>>>>>>>>
>>>>>>>>>>>Is it possible that shutka repeats the same game again and again when
>>>>>>>>>>>chesstiger cannot use learning because of the fact that it is out of book after
>>>>>>>>>>>one or two moves(taking advantage of the fact that tiger has no learning by
>>>>>>>>>>>position)?
>>>>>>>>>>>
>>>>>>>>>>>I am interested to see the games because it seems impossible to do it in fast
>>>>>>>>>>>time control without this idea.
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>Here are the last two games, they have been played with the anti-human option
>>>>>>>>>>on.
>>>>>>>>>>First game was lost by tiger on time, tiger was clearly ahead.
>>>>>>>>>
>>>>>>>>>I do not understand the reason that tiger is losing on time.
>>>>>>>>>It should never happen to computers.
>>>>>>>>>
>>>>>>>>>Uri
>>>>>>>>
>>>>>>>>Without a increment the computer has a disadvantage, as human just make sure
>>>>>>>>you always have a few seconds extra time, eventually the computer runs out of
>>>>>>>>time, unless you get checkmated first.
>>>>>>>
>>>>>>>I disagree because I know that crafty never lose on time with games with no
>>>>>>>increasment.
>>>>>>>
>>>>>>> There are all kinds of delays before
>>>>>>>>engine gets a chance to calculates it's move. This may be as much as half a
>>>>>>>>second per move.
>>>>>>>
>>>>>>>I do not understand it.
>>>>>>>I thought that the delay is 0 seconds
>>>>>>>
>>>>>>>If there is a delay of .5 seconds for move before calculating then humans has
>>>>>>>unfair advanatge because they sometime play faster than 0.5 second per move.
>>>>>>>
>>>>>>>Uri
>>>>>>>Uri
>>>>>>
>>>>>>
>>>>>>all of this is programming problems.  The _only_ delay is between the engine
>>>>>>and the interface, assuming the interface is using timeseal.  Timeseal repairs
>>>>>>the delay from interface to server and back.  However, the engine and interface
>>>>>>are two programs that communicate via (typically) a pipe.  Once the interface
>>>>>>gets the move from the server, the time starts.  It is now up to the program
>>>>>>to read the input, act on it, and produce a move.  Remember, "crafty" is 10
>>>>>>years 'behind' the commercial programs, so I see no point in telling them how
>>>>>>to fix such problems.  :)
>>>>>>
>>>>>>But they _can_ be fixed.  Crafty can play a 60 move game in one second if you
>>>>>>want to see something _really_ fast.. :)
>>>>>>
>>>>>>Bob
>>>>>Except in this setup the engine runs at idle priority at a relatively slow pc,
>>>>>so al the GUI updates are handled before engine get a chance at it. Plus
>>>>>anything else that might be going on. The PC is used as mailserver/domain
>>>>>controller, internet gateway etc. I have not actually messured the 1/2 second
>>>>>delay but this is my estimate. But with 90 moves in 180 seconds it is
>>>>>significant.
>>>>>I have looked at some other games of shutka against tiger, it's al the same
>>>>>shutka plays very fast, tiger gets some won position around move 60 then looses
>>>>>on time.
>>>>>To me it looks meaningless, if I just give tiger engough CPU time the shutka
>>>>>guy is nowhere !
>>>>>
>>>>>Lex
>>>>
>>>>Running a mail server, or a DNS server, requires essentially no computer
>>>>time.  My xeon in my office is running both of these, and the typical CPU
>>>>utilization for either never exceeds 1% of one cpu.  And this includes sending
>>>>out all the email to the crafty mailing list, and so forth.
>>>>
>>>>As far as shutka goes, why not run tiger at normal priority.  It will _not_
>>>>affect DNS lookups, nor mail transfers in or out.  Those things use so little
>>>>cpu time that they will always have a higher priority than a chess engine that
>>>>is computing steadily...
>>>>
>>>>That is what the O/S is all about..
>>>On average the CPU utilization is less then 1 % for these tasks like mail etc.
>>>But once they happen their importance exceeds the importance of this chessgame
>>>running some test, therefore the chessgame runs at idle priority, still getting
>>>most of the CPU, unless real work needs to be done.
>>
>>Sorry, but you aren't thinking about this correctly.  DNS is _not_ compute-
>>bound.  It is _massively_ I/O bound.  Either looking up entries in a local
>>file, or else going to a next-level DNS server with network latency delays.
>>Doesn't matter how busy it gets (I run a network with over 200 machines and
>>the DNS server stuff burns a total of about 1 minute per month or so.)  Same
>>thing for mail.  IE unix sendmail, on a machine that has been up for 2.5 days
>>since we upgraded it, we have this:
>>
>>    root   226     1  0   Dec 10 ?        0:07 /usr/lib/sendmail -m -bd -q15m
>>
>>A total of 7 seconds.  On a machine serving all our faculty email, the machine
>>handles all the departmental mailing lists we have, etc.
>>
>>Neither of those will influence the chess engine in any way.  I run the same
>>processes on my quad and you can bet that if I saw any degradation of any kind,
>>they would be terminated instantly.
>>
>>I/O bound processes simply don't hurt at all.
>>
>
>It's not only I/O processes, I am sure for just I/O you are right, but if
>I want to do some work on the PC, having a chessprogram running at normal
>priority it make the machine behave sluggish, remember it is not a fast
>machine and it has only one CPU.
>

Yes.. but you gave examples of things running that are not going to hurt
performance one iota.  Because they don't need cpu time.  If you run a
cpu-bound process, run it at a _higher_ than normal priority.  Or do just
as you are now, knowing that the DNS/sendmail stuff is _not_ hurting the
chess program even though it is running at a background priority.  That was
the point.  If you think that DNS/sendmail is hurting the program's performance,
you are mistaken.  Whether it runs at normal priority or at a very low priority
is totally unimportant, because those tasks require no cpu time at all.

Now if you run compute-bound stuff yourself, then yes it hurts.  This is how
"scrappy" runs _all_ the time on FICS, and it is how Crafty runs all the time
on ICC.  If something else compute-bound comes in, that gets the CPU, displacing
crafty almost completely.





>>
>>
>>>No matter what OS u use mail tranfers, DNS lookup still have to compete for CPU
>>>time, and for me they are more importand than this chess game. So I won't change
>>>priorities.
>>
>>That's up to you.  Wrong however.  As a process sits blocked on a socket waiting
>>for a DNS query, its priority goes up.  As a process sits computing (the chess
>>engine) its priority goes down.  When a DNS query comes in, it gets handled
>>_instantly_ because the process is no longer blocked and its priority is up.
>>
>>That is why _nobody_ fiddles with priorities on servers. The O/S can handle it
>>far better than you can...
>>
>>
>>
>>
>>
>>>I may run tiger on an other CPU some day, but so far results show me it beats
>>>90 % of it's opposition blind folded and hands tight on it's back :)
>>
>>
>>Stop playing all those 1700-1800's and move up to the big boys.. :)  Try (on
>>ICC) vic11, cptnbluebear, bani, dlugy, any of the others at the top of the
>>rating list. THose make more interesting games...
>
>I would love to, but I don't have an ICC account.
>
>Lex



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.