[Babel-users] Restarting MeshPoint – seeking advice on routing for crisis/disaster scenarios

[Juliusz Chroboczek]

> Have you, or anyone else on the list, worked with meshnet lab before?
> I would be interested to hear how well it matched real world behaviour
> in your experience.

We tried a number of simulation frameworks, then ended up writing our own
scripts that set up containers (using the low-level lxc utilities) and
simulated a network on that.

The main issue was that we didn't properly simulate radio propagation, so
we need to test IRL in order to find out whether link estimation works
well.  And without good link estimation, nothing works.

> I also wanted to ask one more general question about Babel itself, more
> about process than implementation details. When you first started
> developing Babel, was it driven mainly by theoretical reasoning at the
> beginning, with real world testing coming later, or were simulators, lab
> setups, or live networks involved from early on?

It started with the theory.  I was teaching routing, and decided to
implement a routing protocol from scratch, since I like to understand what
I'm teaching.  On Friday evening I started implementing RIPng; by Saturday
morning, I had read a number of papers about RIP, IGRP, EIGRP, DSDV, and
had the impression that I could do better than either of those by
combining the good ideas from them.

For the record, the good ideas in Babel mostly come from EIGRP and DSDV,
plus a number of original ideas.

> Or was it always a mix of both. I am asking from an operator
> perspective, since in my experience many issues only show up once you
> put protocols into messy, real topologies.

I had read a lot of the available litterature, so I was well aware of what
could go wrong.  The main thing that tends to differ between theory and
practice is link estimation, but I was aware of the work on ETX, so I used
that.  (I later tried to improve on ETX, but never managed to convincingly
show that my fancy algorithm actually yielded better routes than ETX in
practical networks.  See draft-chroboczek-babel-diversity-routing for details.)

The other thing that we found out is that 802.11 ad-hoc mode doesn't scale
(it's not designed to scale, and it doesn't), so all production Babel
networks ended up being either wireless or use infrastructure mode.  I am
not aware of Babel ever working well in pure wireless meshes, since we
never found a symmetric link layer that scales well enough.

You may find RFC 8965 useful.

-- Juliusz