Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Junior 8 / Junior 8 CPU Affinity Bug!

Author: Robert Hyatt

Date: 17:37:53 06/05/03

Go up one level in this thread


On June 05, 2003 at 19:21:46, Matthew White wrote:

>On June 05, 2003 at 16:02:51, Robert Hyatt wrote:
>
>>On June 04, 2003 at 23:08:15, Matthew White wrote:
>>
>>>On June 04, 2003 at 22:24:59, Robert Hyatt wrote:
>>>
>>>>On June 04, 2003 at 18:32:43, Amir Ban wrote:
>>>>
>>>>>On June 04, 2003 at 14:54:30, Anson T J wrote:
>>>>>
>>>>>>Has anyone noticed if you do engine matches with Junior 8 / Deep Junior 8, ZX,
>>>>>>YY (any of them). If you set 1 thread per engine (on a dual machine), they share
>>>>>>1 cpu with a net cpu usage of 50% !!
>>>>>>
>>>>>>Deep Fritz 7, Shredder 7 / 7.04 and all other SMP engines I know of don't have
>>>>>>this problem.
>>>>>>
>>>>>>I guess the Junior 8 engines fix the affinity of the single cpu setup meaning in
>>>>>>an engine match or tournament they share the same cpu instead of using one each.
>>>>>>
>>>>>
>>>>>Didn't think of this, damn.
>>>>>
>>>>>
>>>>>>Any fixes known?
>>>>>>
>>>>>
>>>>>The problem should come up whenever running one DJ version against another with
>>>>>no. of threads set to less than no. of processors. I can't think of any
>>>>>workaround. I should get rid of processor affinity in that case.
>>>>>
>>>>>Amir
>>>>
>>>>
>>>>What's the reason for doing this?  The O/S should handle this by itself and
>>>>avoid flipping a single process back and forth between physical processors,
>>>>without the program itself being involved in the decision...
>>>
>>>On a Dual-proc HT enabled system, you would want to make sure that your programs
>>>were getting 1 "real" CPU each and one "virtual one". I'm not sure that the OS
>>>would necessarily behave like that in every case.
>>>
>>>Matt
>>
>>Actually you would want two _real_ cpus, not one real and one virtual.
>>
>>But, in any case, this can't be the issue because how do you _know_ which
>>processor is real and which is virtual, when all you know is 0-n???
>
>Lower numbers are real, I believe. The point that I was trying to make before is
>that rather than giving one engine two real cpus and one engine two virtual
>ones, you would want to partition the real and virtual ones equally. Bear in
>mind I was merely speculating as to Amir's motives, I don't know for sure. I can
>see the reasoning behind your point as well.
>
>Regards,
>Matt


I suspect that with Linux, 0 and 1 are two logical processors on one real
cpu, same for 2 and 3.  That was certainly true a while back but I have not
confirmed it on the latest redhat 9 release.

But regardless, it is a messy thing to deal with in a program and, in fact,
something that probably should not be addressed as it will crush portability.



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.