Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Cluster Toga????
1 2 Previous Next  
Parent - - By Jessica (**) Date 2007-12-30 04:27
I think that if it is a clone then that is REALLY messed up Vas has worked his butt on his work.  Besides I think Vas come out on top of it as well.  Vas always has something up his sleeve :-D  But again I don't want to start that thread all over again because then we ill have over 600 replies! Just curious do anyone really read all those post?
Parent - By turbojuice1122 (Gold) Date 2007-12-30 04:58
I don't read all posts on the forum (though I'll take a quick glance at them), but I read all of the posts in the Belka thread.
Parent - By BB (****) Date 2007-12-29 13:44
Does anyone know the precise firepower of this cluster?

Supposedly, it had 88 processors (2.4 Ghz Opterons) for the WCCC. But as Jeroen noted, it (now) also appears to have a very good book. [It was out of book at move 9 in a Be2 Najdorf against Rybka at the WCCC].
Parent - - By BB (****) Date 2007-12-29 10:54
what the hell is " Cluster Toga " and what's it  all about?

Whatever it is, it played rather well with White against Rybka...
Parent - - By Uri Blass (*****) Date 2007-12-29 11:38
Cluster toga is based on the best free source program(toga) and use several computers so it is basically faster toga.

Toga is clearly weaker than rybka on one processor and I believe that the same is for cluster toga.

I did not see here report about the game of cluster toga against rybka.
Where do you see report about the result of this game?

Edit:
Note that
http://www.chessok.com/broadcast/live.html leads to empty board so I cannot see there games of rybka in the tournament

Uri
Parent - - By BB (****) Date 2007-12-29 11:42
Note that http://www.chessok.com/broadcast/live.html leads to empty board

You have to click on 2007 in the lower-left, then Paderborn, then Live. [Zruty pointed this out originally]

Rybka just resigned.
Parent - - By Uri Blass (*****) Date 2007-12-29 11:53
Thanks

I see that it won against rybka
I do not know if it was bad opening by rybka or if rybka blundered.

Uri
Parent - By Nelson Hernandez (Gold) Date 2007-12-29 13:37
Well, the whole problem is that they followed Nick Carlin's line for 21 moves, then inexplicably deviated.  :)

Nebula,Rybka 2.2 mp (2718) - Flyingfatman,Rybka 2.2 mp (2620) [B90]
16m + 0s, rated, 20.11.2006

1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 a6 6.Be3 e5 7.Nb3 Be6 8.f3 Be7 9.Qd2 0-0 10.0-0-0 Nbd7 11.g4 Qc7 12.Kb1 b5 13.g5 Nh5 14.Nd5 Bxd5 15.exd5 Nb6 16.Na5 Nxd5 17.Qxd5 Qxa5 18.Bd3 Qd8 19.h4 Qd7 20.Rhe1 g6 21.Be4 Rad8 22.a3 Kg7 23.Qd2 f6 24.Rg1 Qc7 25.Bd5 Rc8 26.Rg4 f5 27.Rg2 Rfe8 28.Qd3 Bd8 29.Qb3 Qd7 30.Qb4 a5 31.Qd2 Qc7 32.Re2 Qb8 33.Qd3 Re7 34.Bd2 b4 35.axb4 axb4 36.Bb3 Ba5 37.Qd5 Rc5 38.Qd3 Bc7 39.Ba2 d5 40.Be3 Ra5 41.Bxd5 Bb6 42.Bxb6 Qxb6 43.Bb3 Qa7 44.c3 bxc3 45.Kc2 cxb2 46.Qd8 b1Q+ 47.Kxb1 Ra1+ 48.Kb2 Qa3+ 49.Kc2 Ra2+ 50.Bxa2 Qxa2+ 51.Kc1 Qa1+ 52.Kc2 Qa2+ 53.Kc1 Qa1+ 54.Kc2 Qa2+ ½-½
Parent - - By BB (****) Date 2007-12-29 11:51
Cluster toga is based on the best free source program(toga) and use several computers so it is basically faster toga.

One of the Toga beta versions should allow multi-processing, and Glaurung (not much worse than Toga) does already. I would guess this would be (at least slightly) more efficient than a UCI-based approach (as was claimed about Gridchess, and I'd assume with Cluster Toga also), though if the cluster does not have the desired shared memory model, it might require nontrivial hackery to try to minimise processor wastage.
Parent - - By Vempele (Silver) Date 2007-12-29 12:24 Edited 2007-12-29 12:27

> I would guess this would be (at least slightly) more efficient than a UCI-based approach (as was claimed about Gridchess, and I'd assume with Cluster Toga also)


I still haven't seen anything newer than 2005 to back that up.
Parent - - By Banned for Life (Gold) Date 2007-12-29 23:25
The main selling point for GridChess was its ability to use UCI engines with NO CHANGES. Its unlikely that that has changed.

Alan
Parent - - By Roland Rösler (****) Date 2007-12-30 00:01
I don´t think you are right. For implementing the Young Brothers Wait Concept (YBWC) you have to trim the source code.
Parent - - By Banned for Life (Gold) Date 2007-12-30 00:12
Kais makes it perfectly clear in his GridChess writeup that one of his primary goals was to assemble a cluster computer using unaltered UCI engines (and he emphasizes this point so there is no misunderstanding). The approach is based on what he calls "optimistic pondering". I make no claims as to whether his concept really conforms to YBWC or not, but I see no reason why Kais would try to deceive people about this fundamental point.

Alan
Parent - - By Roland Rösler (****) Date 2007-12-30 00:45 Edited 2007-12-30 00:48
Okay, maybe optimistic pondering works with not parallelizied engines (like Fruit 2.1 or Toga). But I think, the used UCI engines must be parallelizied with wich method ever (YBWC by example).
Parent - By Vasik Rajlich (Silver) Date 2007-12-30 12:10
AFAIK, Kai added YBW to Toga and then created his own cluster controller based on his concept of 'optimistic pondering'. The first step requires Toga sources, the second just requires UCI engines (in this case, multi-processor Togas).

Vas
Parent - - By Roland Rösler (****) Date 2007-12-30 01:15
Appendix: Give me a hint, why Kai doesn´t play with Rybka mp or Zappa and "optimistic pondering" in Paderborn. Why always this bullshit with Crafty, Fruit and Toga (GPL sources)?
Parent - - By Banned for Life (Gold) Date 2007-12-30 06:55
- He received permission from Fabian and Thomas to use Toga for the tournament and they were not planning to attend anyway so there was no conflict.
- He could not play using Rybka unless Vas withdrew and gave him permission (ha ha fat chance).
- He could use Zappa if Anthony would give him permission. This would probably lead to better results but it seems that Anthony does not want to have Zappa play in these types of competitions at this point (maybe he is hoping for a match with some kind of prize as a follow up to Mexico?).

If you remove all the engines that wouldn't give permission or were planning on attending the event themselves, Toga isn't such a bad choice.

Crafty wasn't used as an engine. It was used as the controller for the Toga engines. This required access to the source code. As far as the engine was concerned, any engine with a properly implemented UCI interface should have been ok (but many UCI engines are somewhat deficient in following the spec).

Regards,
Alan
Parent - By Vasik Rajlich (Silver) Date 2007-12-30 12:05
Indeed, Cluster Toga is a legal and fully legitimate cooperation between Fabien, Thomas and Kai. This is the good side of open source.

Vas
Parent - By BB (****) Date 2007-12-30 08:29
Give me a hint, why Kai doesn´t play with Rybka mp or Zappa and "optimistic pondering" in Paderborn.

One word: PERMISSION.
Parent - - By skulibj (*) Date 2007-12-29 12:10
Are you guys telling me that there is no serious programming going on with Cluster Toga? I would think they are consentrating on making the program good for clustering and I would think that is not trivial but rather interresting.
Parent - - By BB (****) Date 2007-12-29 12:32
Are you guys telling me that there is no serious programming going on with Cluster Toga?

I know of no evidence one way or the other - perhaps Kai's Doktorarbeit will elucidate the matter. The typical buzz about Gridchess was that it largely just used out-of-the-box Toga, with some elements of Crafty (for control near the root?), the latter causing questions about its ability to enter the WCCC. The main algorithmic idea appears to be something dubbed "optimistic pondering". There was a thread about it, including a link to a research blurb. The line (quoted in bold by Alan):

> A special advantage of this concept is the possibility to use existing concrete chess programs (so called chess engines) as "workers" (e.g. CRAFTY, DEEP SHREDDER) without the necessity of any engine modification.


seems to imply that the "so called chess engines" are used without much (if any) modification. The "serious programming" appears to be at the meta-level, rather than at the level of engines. It should be noted that many are of the opinion that this is a desired feature, as it will allow other programmes to benefit in such a plug-and-play manner with minimal hassle.
Parent - - By skulibj (*) Date 2007-12-29 13:13
That sounds more likely.
Parent - - By Banned for Life (Gold) Date 2007-12-29 23:40
Conceptually, the method isn't complicated. If you think of running, say three tiers of engines, where there is one engine doing multi PV at the root (pick N = 3 as an example), then having each of the top N moves assigned an engine to look at the next ply, and then another set of engines to look at the ply after that. You would require 1 + N + N^2  engines and each time you got a ponder hit, the engines that were in the correct path would continue processing without interuption, while the ones that weren't ((N-1)/N for a ponder hit and all processors for a ponder miss) would be redirected to provide the additional engines for the last tier.

This is clearly not optimal, but it does allow for easy scaling to a very large number of cores, while current engines seem to hit a wall at 4, 8, or maybe 16 cores.

Regards,
Alan
Parent - - By RFK (Gold) Date 2007-12-30 00:36
If you were to reduce competition to marketable Chess engines, then "Cluster-Toga" would obviously not fit the mold.
Parent - - By Banned for Life (Gold) Date 2007-12-30 00:45
I'm not so sure. Properly marketed, its possible that this could be the basis for the next Hydra.
Parent - By RFK (Gold) Date 2007-12-30 01:02
Well, then that does sort of put an end to this discussion. If " Cluster-Toga" is marketable-as is, then it deserves its position as a viable competitor. I was under the impression that " Cluster-Toga" was in essence an ad-hoc chess entity thrown together specific for this match.
Parent - By NATIONAL12 (Gold) Date 2007-12-30 00:48
robert we lost. end of story.be honest if we had won easily you would not have made that post.the point is we should not have been beaten by a hotpotch.
Parent - By BB (****) Date 2007-12-30 08:41
This is clearly not optimal

Indeed, I would expect a Doktorarbeit to discuss this in more detail. I don't know if informatics (European word for computer science) tries to put itself in the Faculty of Science, or is generally content with an Engineering label, but in the former case, the tournament results of the cluster would not be of such import. Indeed, from the science viewpoint, there should be some "scientific method" followed, and one would expect to find (say): how much gain [in some metric] optimisitic pondering achieves with N processors for N=8,16,32,64, and perhaps compare this to the gain that is achieved from more engine-specific methods. But then, I'm not on Kai's doctoral committee. :)
Parent - By RFK (Gold) Date 2007-12-29 23:41
Because of the divergence this Thread has failed to find the answer to the meaning behind " CLUSTER-TOGA". 

We can only hope and pray that this,  " Lady Ragnell "  will find her,  "Sir Gawain " or, we are doomed to turbulence of the Hag!
Up Topic Rybka Support & Discussion / Rybka Discussion / Cluster Toga????
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill