[Babel-users] How to make babeld updates faster

[Juliusz Chroboczek]

> I need Babel to be able to update nodes at least 2 hops away of a missing node
> in less than 1 second. Is it possible? How do I do it?

> I'm already running babeld with the folowing options;  -h 0.5 -H 0.5

What happens if you reduce the smoothing time constant (-M)?

What happens if you reduce the hello interval further (you can go down
to 0.02, although I wouldn't recommend anything below 0.05).

> But when a node disappears the information takes aroud five seconds

The issue here is that a single lost hello doesn't indicate the node
is dead -- it merely indicates that we lost a packet.  Babel will wait
up to 10 lost packets before it gives up on a node.  In the "good"
case, the other node is strong enough so that doesn't happen, but if
the new node is not very good, ETX and smoothing will delay the
switching.  5 seconds means 3 seconds for ETX and 2 seconds for
smoothing, which is what I'd expect with a hello interval of 0.5.

> A  <=> B <=> C 
> 
> when C dissapears, A takes around 5 seconds to notice it.  Although if the C
> node comes back and connects to B, A detects it pretty fast.

Yes, that's quite typical of ETX convergence.

-- Juliusz