[MLB-WIRELESS] Thanks for the info

[Michele Favara Pedarsi]

2011/3/5 Albert Batiste <arbolito at gmail.com>:
> Thanks guys, I really appreciate your answers! Another question: have you
> ever tried other protocols like OLSR, AODV, B.A.T.M.A.N., Babel...?

I've been playing with wifi mesh networks since one guy in the IEEE
working group list in 2001 asked for feasability of mesh networking
using wifi... then Locustworld appeared (AODV), then Freifunk (OLSR).
I used OLSR since its first draft and I started my own prototype in
Montecompatri (Rome) based on Freifunk OpenWrt customization (Berlin),
in 2005. I rewrote their olsr custom interface using OpenWrt webif
(haserl based) instead of their bash scripts, and added QoS stuff
(integrated 802.11e, enabled ecn, used hsfc+sfq qdiscs; this stuff
should still be in qos-scripts openwrt package).

I've also been experiencing BATMAN and Netsukuku at their early
stages; but too buggy at that time.
I've seen them all (OLSR, BATMAN, Babel) in action last may at Battle
Of The Mesh v3 (Bracciano, Rome); wireless communities gathered under
the direction of local community (Ninux.org), realized a testbed
fonera-based into the camping spot, and tested latest code checkouts
against each other for performances measurements. For detailed results
(cpu/memory, latencies, troughtput) ask some european communities;
there are no large deployment of those protocols so... the only data
around are produced during those grassroots events.

My two cents: they all work just fine as far as you distribute all
resources/services around the network (ie: keep the p2p nature in all
the networking layers) and keep the network small (ie: to make it
scale you need to make multiple mesh layers AND/OR use some
backhauling long distance links); performances vary depending on what
you are searching for (troughput, delays, spontaneous growth).
Coupling those protocols with IPv6 and UDP-only services could make
them work smooth; as well as you need to carry quality measures across
the networking layers (ex: inject radio snr into the ETX metrics of
OLSR).
And... of course... it could be pretty hard to enforce qos or
perimeter security on such network: all nodes MUST follow best
practices AND you CAN'T oblige any of them to do so; as well as you
have the spanning tree problem: can't identify the attacker once it
has gone, unless you apply some kind of cryptography (like x509)
and/or segment properly the network.
In short: an highly disciplined population of network users could
implement a mesh network with great satisfaction in term of
performances but none knows where to find this population :)

Regards,

Michele