Author: Keith Evans
Date: 17:02:50 03/29/04
Go up one level in this thread
On March 29, 2004 at 03:34:23, Steven Edwards wrote: >On March 28, 2004 at 17:44:26, Keith Evans wrote: >>On March 28, 2004 at 15:55:36, Steven Edwards wrote: >>>On March 27, 2004 at 22:49:01, Keith Evans wrote: >>>>On March 25, 2004 at 22:15:16, Steven Edwards wrote: >>>> >>>>>Also, I've got a backup development plan that also uses Lisp and a low NPS, >>>>>whole tree approach. This alternative doesn't rely much on patterns and >>>>>planning, but on a market simulation (!) idea. Here, each node has an instance >>>>>of an interpreter running a program in a Lisp-ish language called NodeScript and >>>>>these instances compete for resource allocation (i.e., greater proportion of >>>>>interpreter step cycles). All the NodeScript interpreter instances run at the >>>>>same time, communicate via messaging plus blackboards, and together perform a >>>>>planless search where the final move selection is reached by consensus. >>>>> >>>>>My NodeScript idea is certainly not like any other chess program known to me, >>>>>and it's also rather unlike the reasoning process of a human player. But it >>>>>does have some similarities to human group behavior, perhaps like a team of >>>>>investment analysts, where economic projections and results guide resource >>>>>allocation and target areas of market expansion. >>> >>>>How many nodes do you think would be running simultaneously? >>> >>>Thousands at least; the only limitation is the addressing space. All nodes run >>>the same uniquely stored NodeScript program; each node only needs to store its >>>own copy of the interpreter state and this is likely under 8 KByte or so. >>> >>>>I don't really >>>>"get" this idea, but it's sort of interesting to me because I could see where >>>>you could implement many of these node processors on an FPGA board, and they >>>>could really run in parallel. (I mention an FPGA board only because it would >>>>make development easier, there would obviously be many ways to approach this >>>>problem.) >>> >>>While a multiple programmable gate array technique may be possible, it may not >>>be the best approach for the above due to the ensuing high shared memory >>>bandwidth requirements. >> >>Does every node need to communicate with every other node, or just a very >>limited subset? >> >>Any hint about what type of messages the nodes would be sending? Just a simple >>example of the type of transaction? > >It would be difficult to give precise answers here as too much of NodeScript is >still only a techical outline. I'll post more details when/if they become >available. Have you seen the papers on turbo codes?
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.