[babel] Diversity routing: a post mortem

[Juliusz Chroboczek <[e-mail]>]

Dear all,

I have just removed support for draft-chroboczek-babel-diversity-routing,
also known as Babel-Z, from babeld.  The sub-TLV numbers remain allocated,
but the code is gone.

The extension was based on an idea by Benjamin Henrion (Zoobab to his
friends), and aimed to choose routes that used as many distinct WiFi
channels as possible, with the goal of minimising self-interference.
While the extension had some interesting theoretical properties (it is one
of the only two known cases of non-isotonic routing in the Internet), we
never managed to prove that it actually improves performance: in practice,
the packet loss caused by self-interference was enough to cause ETX to
choose non-self-interfering routes by itself, and therefore the routes
chosen by Babel-Z were no better than the routes chosen by plan Babel with
ETX.

Since then, the WiFi landscape has changed considerably.  Modern WiFi
chips are able to use up huge amounts of bandwidth (80MHz, or even
160MHz), which makes the simplistic interference model used by Babel-Z
mostly useless.  Even worse, in most environments there are not enough
independent channels to make diversity routing worthwile.

This is an interesting negative result, and one that I wish I had
published.  I remain grateful to Benjamin for getting me to explore the
issue.

-- Juliusz