Archived

This topic is now archived and is closed to further replies.

Lews

Chess and Distributed Computing

Recommended Posts

Hello, For the past two years or so I''ve been running automated chess clients on the Internet Chess Club (www.chessclub.com, I think) (If you''re a member, my machine''s account is Fimbulwinter). I think it''s a great testbed for experimenting with chess programming techniques. This summer I''m interning at IBM, working in grid computing research, and I had the idea to write a client for the ICC that networked any computers on the internet together to create a formidable chess-playing opponent. IBM''s Deep Blue system, I believe, only used 256 1Ghz Intel processors. There are people in my dorm with 2Ghz machines, and many more across the internet. It wouldn''t take very many networked clients traversing chess trees to compete with Deep Blue in terms of CPU cycles. Enamoured as I am with running the most powerful chess machine on the ICC, maybe I''m glossing over some of the most important technical details, such as parceling out jobs to a network of chess computers and minimizing replication of work in a bandwidth poor environment (compared to Deep Blue''s system bus). What do you guys think? Interesting idea? I''d be curious if anyone has run into people doing research like this online? Anyone know of any websites with articles about doing something like this? I think it would be difficult, at best, to keep the CPU utilization of a network of even 10 machines at 100% working on finding the best move in a particular position. How would you parcel work out? How many clients do you think one would need to network together to overcome the lose of speed due to optimizations such as transposition tables in a stand-alone program? If this post seems a bit disjointed, it''s because I''m working a lot of this stuff out as I write it. I''m still in the germ-of-an-idea phase here.

Share this post


Link to post
Share on other sites
If you''re serious about this, I would recommend asking this question at the Computer Chess Club. There are tons of chess programmers there, many who have PhD''s in parallel search algorithms and similar stuff.

It is an interesting idea. I would be very interested to see how such a system performed against GM''s and other computer programs.

Share this post


Link to post
Share on other sites