[Babel-users] scaling of babeld

[Jochen Demmer]

Hi again,

I've been working on different things in order to setup a dev
environment to develop my project which I call pplznet.
https://wiki.junicast.de/en/pplznet/pplznet

Currently I'm running 60 nodes but I would like to get even bigger in
the future. I'm combining Netbox and Ansible together with my Imp1sh
Ansible Collection and there are also some smaller scripts. My aim is
to fully automate the building of a dev environment.

I've been thinking about the TTL in IPv6 is this might pose a problem
and I think it will. With my 60 hosts env I'm already traversing about
25 hops. Real world envs shall get much bigger than 60 nodes.
This is what the dev env looks like:
https://wiki.junicast.de/en/pplznet/dev-environment/diagram

I fear the 8 Bit counter for the TTL might not be enough. Might this
actually be a reason to even prefer Layer 2 routing algos like batman-
adv over layer 3 based solutions like babeld?

What are your thoughts on this?

Thank you
Jochen Demmer

Am Dienstag, dem 24.05.2022 um 18:58 +0200 schrieb Juliusz Chroboczek:
> > What do you think how well would a dynamic routed environment with
> > babeld scale?
> 
> The Babel protocol has been designed to scale up to hundreds of
> thousands
> of nodes.  The current implementations (babeld and BIRD) have not
> been
> tested with more than a a thousand nodes or so.
> 
> Don't let this worry you, though: should you manage to break babeld
> with
> your network, we'll try to quickly fix the implementation issues if
> you
> help us with debugging.  (For example, Daniel Gröber reported an
> issue
> with Babel authentication in his network on 2 May, he helped us a
> lot, so
> the fix was in both babeld and BIRD on 18 May.)
> 
> (Note however that the size of the network might be limited by the
> hardware: some hardware might be unable to install very large numbers
> of
> routes.  This is a hardware limitation, and does not depend on the
> routing
> protocol.)
> 
> > Do you think such a network could handle up to 1.000 OpenWrt nodes
> > or
> > even more?
> 
> Nexedi are running a commercial 800-node network with babeld.
> 
> > What real world bitrate would I be able to achieve?
> 
> Babel is designed so that the routing daemon doesn't look at data
> packets:
> babeld informs the kernel in which direction it should pass packets
> (by
> installing routes in the routing table), and then gets out of the
> way, so
> the kernel or the hardware can push packets at full speed.
> 
> The appliance that you showed us is supposed to be able to push
> 830kpps
> and 8.5GBps.  Assuming that the marketing literature is not lying,
> you
> should be able to achieve that with Babel.
> 
> -- Juliusz