[CUWiN] FreeNet Router Project

[David Young]

On Thu, Jan 13, 2005 at 04:15:58PM +1100, Dan Flett wrote:
> Hi Ian, thanks for your comments
> 
> > Running my own Locustworld mesh but slightly unhappy about the way the
> > source is
> > controlled etc, I've been looking at the other projects around.
> > 
> > WRT based ones , seasoft... wifidog
> 
> OpenWRT at this point is by far the best third-party firmware - the
> closest to being a true Linux distro.  But it has it's own faults -
> user-unfriendliness being the main problem.
> 
> > PC based ones like cuwin.
> 
> I'm trying to move away from PC-based nodes as PCs make noise and take
> up valuable real-estate in the houses of mainstream users.  Geeks like
> you and me don't mind white boxes humming away behind the couch or in a
> cupboard, but we have to learn to think like non-Geeks. :)  A small,
> silent box that can be easily mounted outside is far more attractive.

At this point I should correct Ian: CUWiN runs on Soekris single-board
computers.  That is the configuration we prefer, both for the reasons
Dan mentions, and more.  Finally, when the right device comes along,
we will depart easily from i386.

Dan, you mention portability and embeddability a lot.  CUWiN is hooked
on NetBSD because it is more highly integrated, better-documented, more
portable, and *far* easier to cross-build than Linux.  It also has a nice
architecture for 802.11 that has a remarkable habit of sprouting new
drivers for cheap, ubiquitous cards. For an example of the cross-build
support, I frequently build NetBSD for PowerPC, 64-bit Sparc, and i386
from the same source tree; it's a simple matter of changing command-line
options.  On virtually any POSIX system, the NetBSD build system compiles
the cross-build tools (gcc and the lot) using the native compiler,
and then it cross-builds the kernel, libraries, daemons, and utilities.
(It even cross-builds X Windows!)  Based on the concerns about the various
Linux solutions that you have aired so far, it sounds to me like NetBSD
has some potential in your embedded application.  Take a close look at it.

It sounds to me like you and CUWiN have similar goals.  CUWiN's approach
to meeting those goals is a different than the rest.  Take a look at
what CUWiN is doing on top of NetBSD, too.  Something that I'm excited to
start this year is that we're bringing all of the 3rd-party programs and
libraries under NetBSD's beautiful build system, so that we can target
new embedded systems and systems-on-a-chip with ease.

> > 2) nodes capable of dynamic meshing and routing
> 
> Not necessarily.  Here in Australia government regulations largely
> prevent us from using the Internet to link fixed nodes - so we are using
> wireless as our fixed-backbone.  So that's where our priorities lie.
> Roaming mobile access is not a priority for us yet.  So we are more
> interested in traditional routing protocols such as BGP and OSPF to
> route on our fixed-wireless networks.

Beware: people have wrangled with Zebra/Quagga OSPF on their wireless
networks, even networks built on the point-to-point model you prefer,
and given up.  It may be that you know a lot more than OSPF than anybody
I know, and you can get it to work where others have failed; just make
sure you know why they failed, first. :-)

I am rather skeptical that BGP will be stable on a wireless net.  Do you
know whether anyone has used it on wireless before?  There's no reason
you can't be the first, I suppose.

Dave

-- 
David Young             OJC Technologies
dyoung at ojctech.com      Urbana, IL * (217) 278-3933