WCCC: schedule, pairings, live broadcast

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

frankp
Posts: 233
Joined: Sun Mar 12, 2006 3:11 pm

Re: WCCC: schedule, pairings, live broadcast

Post by frankp »

bob wrote:
The problem here is that it is _easy_ to write an algorithm that will simply not even produce a move given say 64 cores. And I have seen _many_ (including old versions of my code when I first started the parallel search stuff in Crafty) where 8 processors would actually be worse than just using 4 or 2... That's actually pretty easy to cause.
Cough.... that is about where I am right now, only with 2 versus 4 cores, which is hardly faster. Any general pointers? (I memcpy the whole board structure, which might be one issue.).
bob
Posts: 20943
Joined: Mon Feb 27, 2006 7:30 pm
Location: Birmingham, AL

Re: WCCC: schedule, pairings, live broadcast

Post by bob »

frankp wrote:
bob wrote:
The problem here is that it is _easy_ to write an algorithm that will simply not even produce a move given say 64 cores. And I have seen _many_ (including old versions of my code when I first started the parallel search stuff in Crafty) where 8 processors would actually be worse than just using 4 or 2... That's actually pretty easy to cause.
Cough.... that is about where I am right now, only with 2 versus 4 cores, which is hardly faster. Any general pointers? (I memcpy the whole board structure, which might be one issue.).
Several pointers.

(1) the memcpy() can be expensive, but you can limit the cost by not splitting near the frontier nodes (the point where the q-search starts). The farther away from that boundary you split, the fewer splits you will do. Which means the fewer memcpy()'s you do and the less overhead you have. However, it also means that you will have occasions where you can't split for a bit. So lower overhead, and lower processor utilization. Balancing that is important so that you keep things busy without incurring excessive overhead.

(2) move ordering is important so that you can avoid splitting at a CUT node where the extra effort expended is pointless.

(3) You want to avoid any case where one processor is waiting on another to complete some kind of task. This adds up and is a direct property of Amdahl's law and how the max speedup is limited by the fraction of the code that can't be executed in parallel. When a processor needs to wait for another to finish a search, that processor should be able to dive in and help the other processor finish the tree, rather than sitting around waiting, which will kill performance.
Rémi Coulom
Posts: 438
Joined: Mon Apr 24, 2006 8:06 pm

Re: WCCC: schedule, pairings, live broadcast

Post by Rémi Coulom »

Rémi Coulom wrote:Hi,

I have just entered the schedule and pairings on the web site. Here are all the links you need to follow the tournament:

Main page, with cross table:
http://www.grappa.univ-lille3.fr/icga/t ... php?id=192

Pairings:
http://www.grappa.univ-lille3.fr/icga/r ... nament=192

Live broadcast:
http://www.soloajedrez.com/torneos/pamplona09/

Direct link to the webcam:
http://ams01.egihosting.com/90509

The open tournament will take place on May 13.

Rémi
You can now listen to live comments in spanish by Leontxo García:
http://www.soloajedrez.com/torneos/pamplona09/
(live comments will be today and tomorrow).

Rémi
User avatar
AdminX
Posts: 6363
Joined: Mon Mar 13, 2006 2:34 pm
Location: Acworth, GA

Re: WCCC: schedule, pairings, live broadcast

Post by AdminX »

Cool, Thanks
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
swami
Posts: 6662
Joined: Thu Mar 09, 2006 4:21 am

Re: WCCC: schedule, pairings, live broadcast

Post by swami »

Links to ICGA site don't work. So Could somebody else post the rankings here?
Rémi Coulom
Posts: 438
Joined: Mon Apr 24, 2006 8:06 pm

Re: WCCC: schedule, pairings, live broadcast

Post by Rémi Coulom »

swami wrote:Links to ICGA site don't work. So Could somebody else post the rankings here?
All links are working for me now. Here is the cross-table after round 5:

Code: Select all

                  1 2 3 4 5 6 7 8 9 1
                                    0
   1  Rybka         * * * * 1 1 1 1 1
   2  Junior      *   1 * 1 = * 1 1 *
   3  Jonny       * 0   = * * 1 1 1 *
   4  Shredder    * * =   = = * * 1 1
   5  Hiarcs      * 0 * =   * 1 1 * 1
   6  Deep Sjeng  0 = * = *   = * * 1
   7  The Baron   0 * 0 * 0 =   * * 1
   8  Joker       0 0 0 * 0 * *   = *
   9  Pandix 2009 0 0 0 0 * * * =   *
   10 Equinox     0 * * 0 0 0 0 * *
Rémi