[Babel-users] Kernel metrics/route redistribution

[Will Glynn]

On Oct 20, 2010, at 9:52 AM, Juliusz Chroboczek wrote:

>> Relatedly, has anyone attempted, succeeded, or failed to redistribute
>> Babel into or out of other routing protocols?
> 
> I'm happily redistributing from RIPng into Babel.

Great! Out of curiosity, what routing daemon are you using?

>> Exposing the metrics seems like a good idea in general, and entirely
>> necessary if a Babel network exchanges routes with some other network
>> at more than one point.
> 
> That's where we disagree.  I'm not sure whether it makes sense to map
> metrics from one routing protocol into another one.  I feel it is better
> to redistribute by mapping all routes to a single priority -- in my
> case, I'm making the whole Babel network appear as a single hop to RIP.
> 
> If you have a topology in which you want the metrics from two distinct
> routing protocols to be compared with each other, rather than having one
> routing protocol preferred to the other, then I'm very much interested
> in hearing more.

Yep, that's exactly the scenario I'm considering -- specifically, a traditional (e.g. Cisco) wired network that occupies a geographically area that overlaps a Babel wireless network, with multiple wired <-> wireless boundaries.

> (In OSPF terms, Babel only implements the semantics of Type 2 external
> LSAs, not Type 1.  What you want is the semantics of Type 1 LSAs.)

Hmm.

Some assumptions:
	- The wired network is contiguous.
	- The wireless (Babel) network is a stub connected only to the wired network, though connected at multiple points.
	- The wired network is assumed to be a substantially shorter and better path (dark fiber!) than the wireless links.

Therefore, the goal is to minimize the distance traversed over the wireless (Babel) network for packets going to or coming from the wired network.

Traffic out of Babel can follow a default route exported from the Babel nodes physically connected to the wired network. That would get packets from the over-the-air portion onto fiber over the shortest path possible. Redistributing both 0.0.0.0/0 and ::0/0 at that interface seems like it would take care of that without needing to import routes from any upstream routing protocol.

Traffic from the wired network into the Babel network should ideally follow the shortest Babel path possible. This would require exposing the Babel distances to the wired network, which is why I started by asking about that. I think OSPF type 2 external routes would work fine here, as they'd still win longest-prefix matches, and the closest (according to Babel) wireless-to-wired redistribution point would have a lower cost than the others.

Then again, I'm tired, and haven't tried this yet :-)

--Will Glynn