[MLB-WIRELESS] meshing

[Andrew Harcourt]

> So perhaps we are eventually looking at a two-tier network,
> one tier for simple clients (leaf nodes?)  and one for a
> robust, redundant mesh-like backbone network, as outlined
> in the SeattleWireless article.  This would provide for two
> levels of individual interest and involvement too - and
> investment.

Sounds good. Just to clarify: I wasn't suggesting a single set of backbone
nodes without redundancy - that would be silly. We need to have redundancy
*especially* between backbone nodes - I know if I'm going away for more than
a week there's no way I'd be leaving my wireless gear running - it would
pinpoint my place to any aspiring thief running netstumbler.

> The only immediately obvious advantage that an amateur like me
> can see with a universal mesh network is that every time an
> antenna goes up, the network is extended.  And IIRC redundancy
> builds bandwidth?

Well... mostly yes... For short hops near the edge of the network, that's a
reasonable rule of thumb. Routing information isn't all that expensive to
exchange and if we peer our squid caches it will help even more. For nodes
in the centre of the network, every time a node goes up, overall bandwidth
will increase but unless we have some good load-balancing routers around the
place, odds are that one or two links will still be swamped while others
will be relatively lightly utilised. You can tweak this of course, but given
the chaotic nature of the network and its loading, I wouldn't put too much
faith in it. The other thing is that, during peak times, many (all?) of the
paths through the centre of the network are likely to be swamped regardless
of how good the load balancing of our routers is.

> I imagine that the packet-clog that an individual might
> experience could be weighed against the advantage of being
> able to help their mates further up the valley to join in.

Agreed. And for the first four or five people downstream of me, I'd probably
be very happy to support them on a single link from myself. But if I were in
the middle of the network and forwarding packets for 40 or 50 people all
around myself, I'd probably start to get a bit narky.

> I can just imagine the frustration of someone who has spent a
> largish amount of money under the impression he/she will be
> able to do this, only to find that they can't.

Again, agreed. That's why I'm suggesting that leaf nodes need bugger-all
gear (just an AP and maybe an antenna) to connect to other nodes - so that
the entry price can be reduced. The branch nodes that they connect to could
have a mesh-style topology between a few nearby nodes, but if a packet is
going to go for more than a few hops you probably want to get it onto a
dedicated backbone hop.

Think about our supposed advantages over a modem or DSL line: we have lots
of (symmetrical) bandwidth, low latency and traffic is free. If we have more
than a few intra-mesh hops, our bandwidth gets rapidly reduced, our latency
increases and people will start putting QoS limits on the traffic they
forward. That's probably not the outcome we want.

We have to concede the reality that there will be places where a mesh
topology is inevitable. We're not a large corporation and we don't have the
funds to stick a whole bunch of APs on towers and link them with fibre. Our
goal should therefore be to *minimise* the mesh topology.

When we start getting nodes up and talking, odds are it will be on an ad-hoc
routing basis. When we start seeing congestion, that's the place to start
thinking about backbone nodes, and putting in dedicated links along the
lines of congestion. For example, if A has lots of traffic going to E via B,
C and D, it might be better to put in a direct link between A and E, then
allow B to hit E via A, D to hit A via E and C to go either way.

At this stage, it's just food for thought - but it does bear thinking about.


Regards,
Andrew






--
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