Re: networked processors

[Andrew Holt <andrew.holt@xxxxxxxxxx>]

Well IMHO the is no single topology that fit's all.  What you could do
is have a programmable switching scheme so that you configure pipelines,
n dimensional arrays etc.

Not wanting to get into the warfare but i agree with Stephen Pelc (to a
degree).  Supporting C is not essential, but it's a good idea.  lo' cost
comes from hi' volume.  Supporting (arguably) the most common language
will help with the volume.

Regards Andrew

sagalore wrote:
> 
> I don't want to interrupt the little war going on in here, but I have a few questions to ask.  If you you don't like the MISC processor, or the MISC mailing list, or even the Forth language which is fundamental to the F21 instruction set, then don't subscribe to the MISC list.  It reminds me of christian newsgroups being 90% atheist and atheist newsgroups being 90% christian.  The only people being religious in here are the ones that don't belong here.
> 
> The F21 being such a small chip, I believe it's power is realized in *multiple* F21s being networked and treated as a single processor.  I know there are several topologies that could be used - token ring, parallel, star... what do you think should be used for the best performance gain?  What overhead is involved?
begin:vcard 
n:Holt;Andrew
tel;cell:0802 914670
tel;work:01257 241589
x-mozilla-html:FALSE
url:www.sun.co.uk
org:Sun Microsystems;Sun PS Custom Engineering
version:2.1
email;internet:andrew.holt@uk.sun.com
title:Technical Consultant
adr;quoted-printable:;;City Gate=0D=0ACross Street;Sale;Cheshire;M33 7JR;UK
x-mozilla-cpt:;1
fn:Andrew Holt
end:vcard