[Babel-users] Diversity Routing for the Babel Routing Protocol

[Dave Taht]

On Wed, Jul 2, 2014 at 8:22 AM, Juliusz Chroboczek
<jch at pps.univ-paris-diderot.fr> wrote:
>>>> Regarding Z3 and RTT, you're right, it's not exactly clear.  The current
>>>> implementation of Z3 applies the diversity factor *after* all other costs
>>>> are computed, included the one induced by babel-rtt.
>
>>> Huh?  RTT and Z3 work at completely different places -- RTT tweaks the
>>> cost computation, while Z3 changes the announced metric.
>
>> And we've just demonstrated that it is confusing, even to the authors
>> of the IDs.
>
> Don't blame Baptiste, it is a difficult topic.  There's good reason why
> we're the one of the only production-quality protocols in the world that
> support hybrid metrics.  (There's IGRP, not sure about IS-IS.)  It might
> be interesting to write up everything we've learnt on the subject, but
> that's more work than just an off-the-cuff comment on a mailing list.
>
>> Secondly, I was also hoping for an extended example of why diversity helps
>> on longer paths as per my first email.
>
> I've added a second example.  Is there anything else you had in mind?

Well, showing that performance falls off 1/2,1/3,1/4 (or is it 1/2,
1/4, 1/8?) as the number of interfering links on hops grows by 2,3,4
is a good thing to beat the shortest path ethernet-thinking-only
people over the head with.

Secondly,

"When routing data from A to F, the route through B and C has two
interfering links: two packets sent by A and C at roughly the same
time will collide, and both will need to be resent. Again, no such
issue arises with the route through D and E."

This is true in the case of the hidden node problem, somewhat true in
the case of adhoc, and not true in the case of an AP/sta. In the
latter case especially, packets transiting in each direction are
scheduled, not "colliding". The net loss in throughput is still
enormous, have a paper around here somewhere on that...

While I'm at this, a long standing bug in either the kernel or babel
or both has been in determining the channel in the case of a bridged,
or ap-mode connection.  let me start another thread...


> -- Juliusz



-- 
Dave Täht

NSFW: https://w2.eff.org/Censorship/Internet_censorship_bills/russell_0296_indecent.article