Author: Robert Hyatt
Date: 06:40:35 02/09/01
Go up one level in this thread
On February 09, 2001 at 02:19:47, derrick gatewood wrote: >Distributed computing has been a very exciting field, especially in the past few >years. One night while playing my computer account on >chess.net(Catcher_in_the_Rye) I was closing down my SETI program so that my >computer wouldnt totally slow down, then I had an idea. Wouldnt it be >exciting/great if we could apply the same ideas as SETI and RC5..whatever.. in >the area of computer chess. Now, I am aware there are certain problems that >are involved with dirstributed computing. One, chess relies on the results of >previous searches to begin further searches. This makes this solution less than >ideal. However, I was thinking along the lines of the primary computer doing an >initial search to say even 5 or 6 ply, then have it send data units to the other >computers assigning them a certain area of the search tree that it would like >them to complete. Once they are done with their part of the tree, they can >then send the unit back in the form of a rated result.. the primary computer >can then store the result in its tables and if time permits it could send the >secondary computers even more packets for processing. Now, this is where the >second problem in distribuuted computing comes into play, bandwidth. There >would be a lot of information being sent back in forth and it would be needed in >a timely manner. This would make most dial-up users not be able to participate >because it would bog down their connection and force the primary computer to >slow down and wait to receive the slow packet. But... Lets say that a person >has a server farm of about 15-20 computers all wired with 100mbp/s lan.. This >would make it even more viable? Or would it? This is just an idea, and I >would like to hear what the real computer scientists have to say about it. I am >only a networking guy and know little about actual programming, although I have >tried to write my own chess program from scratch only to get bored of mediocre >results. Thanks in advance for all replies. the idea is definitely interesting. The issue revolves around two measures: (1) time limit for a search; (2) network jitter/lag time; I think blitz is impossible over a WAN, although a LAN-based distributed algorithm could work fine. Longer time controls might do better on a WAN where jitter/lag is a tiny percentage of the total move time...
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.