OSPF network_type optimization for P2P links

[jbemmel]

I noticed that

netlab/netsim/modules/_routing.py

Line 87 in 34c0569

elif len(intf.get('neighbors',[])) == 1:

sets the OSPF network_type to "point-to-point" in some cases. In my case, it was VLANs between 2 switches with no hosts connected to them (yet)

It occurred to me that even if there were hosts, there would still be only 2 OSPF speakers on the segment - so could this optimization be broadened? Instead of checking len(neighbors)==1 filter for neighbors that have the corresponding routing protocol enabled?

[ipspace]

I don't think that's a good idea for scenarios where someone could turn on OSPF (or IS-IS) on another device (IS-IS labs come to mind ;). Anyway, I don't think I've ever seen OSPF/IS-IS P2P type used on edge segments (even though that would technically make sense).

[jbemmel]

That's the thing: With Netlab, you get SVI interfaces with p2p OSPF network_type for VLANs with no hosts. Add hosts, and it becomes a broadcast network

Perhaps instead of checking len(neighbors)==1, we should check for link.type=='p2p'?

[ipspace]

Looks like we have another bikeshedding exercise. What's wrong with the ospf.network_type: broadcast VLAN attribute?

[ipspace]

I have no problem with making OSPF network_type global and node attribute, but don't complain about the results you'll get ;)

[ipspace]

But we could do VLAN groups just like we have node groups:

How about

groups:
  edge:
    type: vlan
    members: [ v1, v2, v3 ]
    ospf.network_type: broadcast

That would be pretty cool, and we could do it for other dict-like objects as well (VRFs come to mind)

[jbemmel]

type could be inferred from the first member, and then check that the other objects are of the same type

[ipspace]

type could be inferred from the first member, and then check that the other objects are of the same type

... and the next microsecond someone would use the same name for a node and a VLAN 😜

[jbemmel]

...and we could check for that too (and call it "AI")