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

[Juliusz Chroboczek]

> If your network consists of pretty static nodes (fixed routers on the roof),
> you can tune your settings to update the routing less frequently.
> 
> Adding and removing node can take time (30 minutes) it does not need to be
> instant.

The problem is what happens when a node breaks.

In general, it's fast to detect that a new node is added (as soon as you
receive a Hello, you know the new node is available) but difficult to
detect that a node has gone away (you need at least three missed Hellos in
order to confidently affirm that a node is gone).  And if you don't react
fast to a node going away, you're going to create a blackhole, which is
annoying for the users.

Babel mitigates this somewhat by having an explicit protocol for leaving
the network (if you kill the babeld process before you disconnect a router
from the network, then it will notify its neighbours within milliseconds),
but that doesn't help when routers crash or are brutally switched off.

So in my opinion, you need Hello intervals on the order of a few seconds
even in a stationary network, since occasional outages of 15s or so are
acceptable, but not more than that.

-- Juliusz