X-Message-Number: 24056
Date: Thu, 6 May 2004 12:36:23 -0400
From: Jeffrey Soreff <>
Subject: Re: CryoNet #24040 - #24046

>If we were to try to actually build a structure with (to be simple)
>1 billion nodes, each one with a connection to every one of the others,
>then it should be clear that we'd run into serious practical problems.
>Each node will have 1 billion - 1 connections emanating from it.

Yes, a full connection of all N processors directly to all N-1 other
processors is quite unreasonable for large N.  There are other simple
topologies which can pass messages between any pair of processors,
albeit through intermediaries, which are physically much more reasonable.
One of the older ones is a hypercube, where the N processors each have
log2(N) connections to other processors and a message can go from any
processor to any other processor in log2(N) steps.  Keeping with the
N = 1 billion case and the port size = 1 nanometer example, this would
have 30 ports per processor, so the 30 nm^2 area of the ports could
occupy the surface of a sphee of radius 1.5 nm.  Given the ~GHz/~kHz
speed ratio of contemporary electronic logic to neural speeds, losing
a factor of 30 to the message passing isn't a large cost compared to
the factor of 10^6 from the technology speedup.

                                Best wishes,
                                -Jeff

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=24056