Why Christophe doesnt answer???

[chesstango]

u kno what?
i guess chess markets are clearing from now on; chessbase is almost out; lokasoft is almost out ; chess tiger for palm is out.....
I should say that Vas, Mark and the author of Shredder uci have the best customers services.

[Eelco de Groot]

Paulo, he answered that question many times. He thinks that it is pointless to add SMP support until he improves his engine enough to be one of the top engines.

I didn´t read nothing about Chess Tiger SMP, thanks, Tony.
But if I asked a question to Christophe two times, why he
didn´t reply? At least he could write: " Don´t bother me"

Paulo Soares

I am sure that one day Chess Tiger will go SMP too Paulo! I can't, and don't want to answer here for Christophe, but I hope he does not mind me doing just some speculative reasoning about the issues involved in making a parallel chessprogram, going by what Christophe and others have written in this forum and elsewhere:

• The most important reason, in my opinion at least, why parallel processing for Chess Tiger will not be realized sometime really soon is probably the reason that Tony mentioned above, that Christophe first wants to perfect the single processor version of Tiger.
• Another complication of matters; at the moment I think Christophe doesn't yet posses a dual core machine or something like that for his programming. At least the last time he wrote about that, I think it was last summer in this forum, he did not have such a machine, which makes it a bit difficult to develop and test any multiprocessing software!
• Also, I think that Chess Tiger does not use totally standard Alpha Beta algorithm, Christophe has written once or twice for instance that Tiger does not use, or suffer from, the concept of "late" moves as in "Late Move Reductions", so that implies Tiger does things rather differently in many respects. That surely further complicates things going parallel, because there is no standard way to go about it, Christophe has to develop this totally by himself.
• Chess Tiger was always developed to be able to run both on small machines and on PCs, it can run on Palms and PDAs with basically the same standard Tiger-program as on a PC. Making a Multiprocessing version of Tiger will mean leaving this path, Christophe has to make and develop at least two different versions of his program as an SMP Tiger will not be much use on a Palm.
• Making an SMP version of a chessprogram, or any program, but especially of the complicated entity that a top chess-program is, is really a very complicated issue. There is no single right way to go about things I think, and when the hardware evolves, the best way to parallelize things also changes with it. There are ways to keep it relatively simple, but that also implies limits in performance when the number of processors increases. As I understand it, at the moment chessprograms use basically three different forms of parallel processing on PC hardware (not counting PC-clusters etc.)
  
  1. The simplest way is to let two, or more, threads analyze the same position, which seems rather pointless at first but the threads can profit from each other's search by sharing their results in a shared hash table. This does not scale well to more than two processors, but can sometimes work relatively well on a dual. A case in point seems to be The Baron from Richard Pijl. On the Winboard Forum there was an interesting discussion lately about using this shared hash tables method, and Richard Pijl described some of his experiences there with it in the SMP Baron. I believe this method is also called the ABDADA algorithm: http://wbforum.vpittlik.org/viewtopic.php?t=6250
  2. A method that scales better seems to be something that is sometimes called the YBW, "Young Brother Wait" algorithm. I don't know much detail but basically the threads divide the work they are doing so they do not all search the same tree. This does scale better than ABDADA but is not yet optimal for a larger number of processors. Programs like Glaurung and I suppose most of the "Deep" programs use it, including Rybka.
  3. A third method scales even better to more processors, as it has more flexibility in where to split the search tree. It is also more complicated to implement I imagine, and not so well documented because there are relatively few programs using it. I think this is called DTS, and Crafty and Zappa are the main chess-programs using DTS, as far as I know. It was described in a research paper by Robert Hyatt in 1994, I suppose it was published in an issue of the ICCA journal from that year but I am not sure about that. There was an interesting discussion, at least for the technically interested people I think, about some of the details of implementing Rybka's search on modern hardware by Vasik on Rybka forum and the possibility of one day implementing DTS in Rybka: http://rybkaforum.net/cgi-bin/rybkaforu ... pl?tid=640 That's where I first read about this DTS algorithm.

I do not know much about either the hard- or software issues but I still hope this provides some background to some of the issues involved in making a chessprogram, and ChessTiger more in particular, go parallel!

Regards, Eelco

[Eelco de Groot]

[*]A third method scales even better to more processors, as it has more flexibility in where to split the search tree. It is also more complicated to implement I imagine, and not so well documented because there are relatively few programs using it. I think this is called DTS, and Crafty and Zappa are the main chess-programs using DTS, as far as I know. It was described in a research paper by Robert Hyatt in 1994, I suppose it was published in an issue of the ICCA journal from that year but I am not sure about that. There was an interesting discussion, at least for the technically interested people I think, about some of the details of implementing Rybka's search on modern hardware by Vasik on Rybka forum and the possibility of one day implementing DTS in Rybka: http://rybkaforum.net/cgi-bin/rybkaforu ... pl?tid=640 That's where I first read about this DTS algorithm.

I found the following link about DTS, Distributed Tree Search, and other approaches to parallel searching:

http://www.cs.cmu.edu/afs/cs/project/ja ... node3.html

[Tony Thomas]

I did so; i sent him a couple of emails but no answer at all; anyway if i were sure that sending him the payment i will receive the key i would do it; but this would be done for sure; and at this moment for sure is to receive a signal from christophe......Im so sorry for those people who manage their businesses this way..... You can see how well Vas or Mark manage their businesses!!!

Oh my god, your complaint doesnt even make any sense. How can you expect him to send you the key before you send him the payment? Ofcourse you will receive your key after you send the payment. I have never in my life received a merchandize before I paid for it.

[PauloSoare]

[*]A third method scales even better to more processors, as it has more flexibility in where to split the search tree. It is also more complicated to implement I imagine, and not so well documented because there are relatively few programs using it. I think this is called DTS, and Crafty and Zappa are the main chess-programs using DTS, as far as I know. It was described in a research paper by Robert Hyatt in 1994, I suppose it was published in an issue of the ICCA journal from that year but I am not sure about that. There was an interesting discussion, at least for the technically interested people I think, about some of the details of implementing Rybka's search on modern hardware by Vasik on Rybka forum and the possibility of one day implementing DTS in Rybka: http://rybkaforum.net/cgi-bin/rybkaforu ... pl?tid=640 That's where I first read about this DTS algorithm.

I found the following link about DTS, Distributed Tree Search, and other approaches to parallel searching:

http://www.cs.cmu.edu/afs/cs/project/ja ... node3.html

Eelco, more one time: You are great !!!
Many thanks, but unhaply I can´t understand about SMP, late move redutions, Alpha Beta, and others, but you explain to me the
probably motives that impede Crhistophe to make a SMP version.

Paulo Soares

[geots]

Hmm. I always ask for the engine first.

[PauloSoare]

I asked him to send me the key for chesstiger 15.2 for my new palm and he didnt even show .....
what's up? Is the customers service not serving anymore?

Maybe you have to ask to Lex.

But.... I asked two times to Christophe about an SMP version and he
didn´t reply.
It´s strange, because in the Rybka forum, Vas reply many questions
made by the costumers, it´s a great forum. Stefan reply all questions
that I made. Gerhard, from Loop, and others, the same.

It´s simple: Christophe, do you think to launch a SMP version?
1- Yes.
2- No.
3- I don´t know yet.
4- Other reply.

I think that no reply it´s not good, because Christophe post in CCC.

Paulo Soares

When you talk about Rybka, the first name that you remember is Vas.
Well, IM Larry Kaufman is doing a great job in Rybka Forum, and I forgot
to write his name, sorry.

Paulo Soares

[Sylwy]

Christophe works to Chess Tiger SMP ! Be quiet please !

[mclane]

u kno what?
i guess chess markets are clearing from now on; chessbase is almost out; lokasoft is almost out ; chess tiger for palm is out.....
I should say that Vas, Mark and the author of Shredder uci have the best customers services.

????

neither is chessbase "almost out" nor is lokasoft. and chess tiger for palm is especially NOT out.
Why do you write those things ?

[mclane]

strange indeed.
he cannot send the key if he does not know the number that your palm creates. than he waits for the payment and THEN sends the key.

tony is right here.