Author: Omid David Tabibi
Date: 05:48:52 09/18/04
Go up one level in this thread
On September 17, 2004 at 18:09:23, Uri Blass wrote: >On September 17, 2004 at 17:49:50, Dann Corbit wrote: > >>On September 17, 2004 at 16:56:52, Gian-Carlo Pascutto wrote: >> >>>Abstract: >>> >>>"In this project it is examined how the use of a specific data structure called >>>a bitboard affects the performance of parallel search." >>> >>>Conclusions: >>> >>>"Our experiments showed that speedup was not near ideal using many processors. >>>Whether or not this was due to the use of bitboards is unclear." >>> >>>So, what was the goal of this research again? :) >>> >>>Kudos for including your source in any case - at least your results can be >>>verified and further investigated, even if you didn't really manage to produce >>>much useful results... >> >>I think everyone is being a little harsh. >> >>Similar to the paper by Marcel V.K., it is an interesting piece of work. It is >>easy to read and understand. >> >>Of those people who have managed to accomplish a parallel implementation of a >>chess engine (I am guessing that there are less than 10 in the world) only a few >>have bothered to explain what they are doing, and only Dr. Hyatt and Mr. >>Rasmussen have given out their source code. > >I think that there are more than 10. >I can easily count more than 10 programs that use parallel search not by >Rasmussen and I do not know all the programs that use parallel search. > >1)Junior >2)Fritz >3)Shredder >4)Diep >5)Sos >6)Falcon >7)Sjeng >8)Amy >9)Crafty >10)Zappa >11)Storm >12)Hydra >> >>Writing a parallel chess engine is not trivial in the least, since it definitely >>requires an understanding of multithreaded programming which is also fairly >>unusual. > >David Omid claimed that it does not require a lot of work and I guess >that I am simply too ignorant not to know the usual knowledge of programmers. "not require a lot of work" is a very relative term. I believe that I have spent much less time in comparison to programs using multithreading. I have all sorts of global variables all over the program, and so going for multithreading would take much longer, to reorganize the code, and then spend a very long time debugging. In multithreading both threads are mapped to the same memory space, and so it is much harder to avoid and detect bugs. In multiprocessing you only share the parts you want via shared memory (e.g., hash table), and so it takes a considerably less time to implement and test, assuming that you haven't designed your program with future multithreading in mind. > >I guess that the reason that most programmers did not write parallel version of >their engine is simply lack of interest and most programmers know more than me. > >Uri
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.