[Babel-users] babel in vyos

[Dave Taht]

On Fri, Apr 28, 2023 at 1:35 PM Juliusz Chroboczek <jch at irif.fr> wrote:
>
> > One place where things fall down also is in ECMP, (which babel does
> > not do presently?)
>
> It does not.
>
> ECMP improves throughput in some cases, but unless done very carefully it
> tends to improve latency (the effective latency of the aggregated link
> tends to be higher than the latencies of either of the underlying links).
> I don't currently have an algorithm that would reliably determine when
> ECMP can be used without impairing latency, but perhaps further work on
> the RTT metric could yield something useful.
>
> (There's this good friend of mine who taught me to disregard throughput,
> and optimise for latency.  His name is Dave.)

I miss ya too buddy!

I am now in a world of mikrotik and ubnt where cake and fq_codel are
becoming universal, and dealing with the kinds of problems that WISPs
deal with every day,
one of them being the inadequacies of olsr in complicated wireless
environments. One libreqos user has 10! mmwave hops that goes to hell
even in a light drizzle... others have 5Ghz fallbacks for links like
that...

Had I known it would have taken 14 years to get to this point, and
years more to *maybe* start making a dent in the routing problem(s), I
would have stayed on my beach in nicaragua, I think.

btw, libreqos is GPLd and there is a cool demo of it here:
https://payne.taht.net. One of the coolest features is a "piano roll"
of real traffic over time, however seeing that against real traffic,
as opposed to the simulation there, is most revealing. Fixing network
after network of 2k+ subscribers in a few minutes has been quite
rewarding and almost makes up for all that time spent...

Anyway, if ubiquity of fq_codel can be assumed, other things seem feasible.

>
> -- Juliusz



-- 
AMA March 31: https://www.broadband.io/c/broadband-grant-events/dave-taht
Dave Täht CEO, TekLibre, LLC