[MLB-WIRELESS] meshing

[David Arnold]

-->"Roger" == Roger Venning <r.venning at telstra.com> writes:

  Roger> I would dispute that ad-hoc mesh networks have 'poorly
  Roger> understood' routing behaviour. The IETF Mobile Ad-hoc
  Roger> Networks working group (MANET
  Roger> http://www.ietf.org/html.charters/manet-charter.html) has
  Roger> been going for many years now, and embodies considerable
  Roger> research in the field.

yep.

  Roger> The particular instance of an ad-hoc network that an
  Roger> operational Melbourne Wireless would represent is almost a
  Roger> trivial case given the distinct lack of mobility and the
  Roger> reasonably large link lifetimes that would be expected. This
  Roger> doesn't mean that the problem is simple though.

when comparing the understanding of fixed, provisioned routing with
mobile, ad-hoc routing, i think it's fair to say that manet/mobihoc
stuff is relatively poorly understood.

not to say that there aren't protocols, and haven't been tests, nor
that there aren't very smart people trying to make it work well, but
there's very limited experience with real deployment that i'm aware
of.

i'd be reluctant to base all our functionality on any of these
protocols, especially given that (as you point out) we don't really
need the mobility that drives them.  something widely and well
understood seems to me more attractive as a starting point.

  Roger> One of the largest issues involves not connectivity - the
  Roger> simplest goal of any routing algorithm that could be run on
  Roger> the network - but rather traffic engineering, the 'art' of
  Roger> getting load balancing across not just multiple links but
  Roger> multiple paths. This is a problem that is critically
  Roger> dependant upon traffic dispersion - where the traffic sources
  Roger> and sinks are. It will be a big issue if there is for
  Roger> instance one particularly popular resource within the network
  Roger> - e.g. a Internet connection, wormhole to another wireless
  Roger> network, etc.

yep.

  Roger> One interesting thing to note about 802.11 is that the medium
  Roger> access control layer is not optimal for supporting
  Roger> peer-to-peer forwarding as found in a mesh network
  Roger> situation. For more information have a look at a paper that
  Roger> was published last year in the IEEE Communications Magazine -
  Roger> http://www.cs.ucsb.edu/~cs290i_mc/papers/80211_adhoc.pdf.

reading ...

  Roger> I've been thinking for a while now about the need to
  Roger> bootstrap ourselves up from a very low density network
  Roger> towards a network of a 1000-10000 nodes. I think it is
  Roger> essential that the solution that serves the network to begin
  Roger> with be low cost - just one wireless interface. I am almost
  Roger> certain though that any architecture that serves the low
  Roger> density starting point will be adequate for a 'successful'
  Roger> Melbourne Wireless. However we must bridge that river when we
  Roger> come to it. 

i'm not aware of a routing protocol that would cope, but ... i'd be
interested to hear suggestions.  the principle is certainly nice, if
the practice can be made workable.

  Roger> I'll 'publish' my thoughts & software kit on how to drive
  Roger> either an access point or a wireless interface from a PC
  Roger> class router node in a way that provides service to local
  Roger> wireless clients as well as supporting connectivity to the
  Roger> mesh Real Soon Now (TM).

now *that* sounds promising ;-)

care to share any early-access details?






d

--
To unsubscribe, send mail to minordomo at melbwireless.dyndns.org with a subject of 'unsubscribe melbwireless'  
Archive at: http://melbwireless.dyndns.org/cgi-bin/minorweb.pl?A=LIST&L=melbwireless
IRC at: au.austnet.org #melb-wireless