Author: Simon Waters
Date: 17:14:23 10/30/00
Go up one level in this thread
On October 29, 2000 at 01:41:31, Ratko V Tomic wrote: >> But clusters can work. > >AOL, with its 25 millions subscribers could yield a good size "cluster" with >almost no network hopping. With the right diagnostic and configuration tools >one could extract thousands of clusters (and compute routing tables) with >intra-cluster packet delays (with high probability) below 50ms or even 30ms. I can see it now - Extra terrestrials versus Seti@Home.... Connection speed would be an issue - the Academic networks would probably be a better bet - 100Mbps LAN's interconnected at 155Mbps (at least over here). Over a typical dial-up line it takes tens of milliseconds to send and receive packets (That is a long time these days - honest) - and you have to figure out how to avoid doing this as much as possible (Which is what Bob is looking forward to thinking about I expect). Interesting and revealing is how much of the big "distributed projects" actually get done on otherwise idle supercomputer installations - the big interconnect speeds on the massively parallel machines is a big plus. >And if one is looking only to create and then have room to store a gigantic >table-base (say, as a worldwide endgame super-oracle), connection speed wouldn't >matter. Similarly, the projects of the type 'Crafty goes deep' would work quite >well with any number of computers via internet (and then 'Crafy goes dot-com' >could nail for good the controversy on 'diminishing returns in chess'). Hmm - interesting idea which may scale better - I suspect it has it's own issues like indexing - "I just want to look up this positions in the database of all 10 piece endings to see who has won.... Who has this bit of the database... Has it been calculated yet.... ".
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.