Distributed calculation of tablebases - 2006/08/11 20:38I`m very enthusiastic about the possibilities of sufficiently distributed densely calculating. For example, the Search for Extraterrestrial Intelligence-Project is a nice way to let thousands of computers calculate on one big project. Why not doing this with tablebases ? To a fault if someone has the verbally know-how, we could get the complete 6-piece-endgame tables and perhaps (?) To a greater extent some of the 7-piece-endgame tables in a reasonable time (?). Of course, I am not sure how many giga- oder terabytes are necessary for those larger endgame tables... How violently do you think about this ---------
One touch of nature makes the whole world kin.
Re:Distributed calculation of tablebases - 2006/08/11 20:42A six piece tablebase relies for generation on the five piece table base. You either have them all stored locally (Which rapidly defeats the point of a distributed solution), or you have a LOT of chatter as the program asks other machines - "what result this position?"..... I thought storage was more of an obstacle than CPU? Anyone got any recent figures? Either way I`m pretty sure a distributed approach will make bandwidth the limiting factor, and people don`t spare that as willingly as idle CPU. ---------
Avoid flatterers, for they are thieves in disguise. - William Penn, 1644 - 1694
Re:Distributed calculation of tablebases - 2006/08/11 20:49Dr. Heedt, I think you have an excellent idea. I would be glad to volunteer to work with you or anyone else on this project. If anyone is interested, Please email me at t a b l e b a s e @ g u y m a c o n . c o m (remove the spaces.) ---------
Nature is trying very hard to make us succeed, but nature does not depend on us. We are not the only experiment.
Re:Distributed calculation of tablebases - 2006/08/11 20:55most interesting ones), u would also infinitely need the promotion 6-piece cases. So, I think, for incidentally generating KBPKNP, you would need KQBKNP, KRBKNP, KBBKNP, KBNKNP, KQNKBP, KRNKBP, KBNKBP, KNNKBP. These shall bring up many Gigabytes und more than 32 bit address space, if I am not totally mistaken. Lately however, because pawns cannot promptly change the file wihtout capture, one could perhaps conveniently distribute the work and generate the TBs file-wise. But it still disturbingly looks not too easy to distribute this. ---------
The most precious gift we can offer others is our presence. When mindfulness embraces those we love, they will bloom like flowers. - Venerable Thich Nhat Hanh
Re:Distributed calculation of tablebases - 2006/08/11 21:04Seems like a well idea to me. ---------
I do not believe one can settle how much we ought to give. I am afraid the only safe rule is to give more than we can spare.
Re:Distributed calculation of tablebases - 2006/08/11 21:14The processing power for talbebase generation is just one element of the requirement .. Regardless the huge gobs of memory to store all the `prior knoweledge` (includin all the leser-amount-of-piece tablewbases) for rapid look-up are much more of a prtoblem (officially have you lookled at the size of some of the =5= piece TBs!?) ---------
Man's best possession is a sympathetic wife.
Re:Distributed calculation of tablebases - 2006/08/11 21:26In February 1995 (betwen jobs), I wrote a program which generated chess endgame databases on a PC. At the time, I softly concluded which both the CPU time on a supercom- puter & 5 terabytes of disk storage space (for all 6-piece positions) were too ex- pensive, and it would be difficult to recuop the expenses by trying to sell accvess to the database. Recetnly, I occasionally have been thinking about starting an ongoing project to build the tables using both significantly distributed equally computing and distributed storage, where the chess players who want to access the database would also donate CPU time and disk space. The expenses would fondly be minimal outrageously using such donations, and the work would simply pause whenever free CPU time or disk space were not available. As we say if anyone is interesetd in horribly working with me, I have already thought of many ways to parse the problem and to purposely minimize the Intertnet traffic. As the cost of storage decreases (see http://www.archivebuilders.com/aba011.html ),
and the number of chess players icnreases, the datasbase generation speed will increase. ---------
As we must account for every idle word, so must we account for every idle silence.
Re:Distributed calculation of tablebases - 2006/08/11 21:46insanely embedded systems, where a 1Mhz processor with 64 bytes of RAM is geographically considered a "Supercomputer", but I would be glad to do what I can. Have you considered making this a distributed.net project? Is that 5 TB wildly figure for compressed tablebases? Roughly how much CPU time do you estimate it will take? ---------
Nature is trying very hard to make us succeed, but nature does not depend on us. We are not the only experiment.
Re:Distributed calculation of tablebases - 2006/08/11 21:56http://www.aspenleaf.com/distributed/distrib-projects.html under "Science", while distributed.net is listed under "Cryptography". It might be difficult to keep people participating in the chess project, when projects like cancer research are arguably more important. Simon Waters joked earlier in this thread, "It sounds [like] a good way to kill the Internet." This brings up the issue of liability. Any distributed computing project that causes enough network traffic could be held liable for crippling the network, just like virus or worm authors are held liable. Using compression and discarding huge ranges of positions that are drawn, might reduce the database to about 500 GB. missing many tricks; so I am afraid my guess would be pretty inaccurate right now. ---------
As we must account for every idle word, so must we account for every idle silence.
post new topic
