[c-nsp] filter LDP bindings

Oliver Boehmer (oboehmer) oboehmer at cisco.com
Thu Aug 14 03:41:41 EDT 2008


Saku Ytti <mailto:saku+cisco-nsp at ytti.fi> wrote on Thursday, August 14, 2008 8:17 AM:

> On (2008-08-13 20:38 +0200), Oliver Boehmer (oboehmer) wrote:
> 
>> well, this dependency on what other LDP neighbors send is not really
>> in-line with the independent control mode LDP operates in, so the
>> implementation might not be straight-forward.
> 
> I think we have misunderstanding here. All boxes would 'stupidly'
> accept and readvertise everything they get, no additional states
> here, plain 'ol ios behaviour without LDP ACL.

Well, I think this is the catch: In independent control mode, LDP does not "re-advertise" something like a distance/path-vector routing protocol does, it advertises its local bindings. So to implement a "re-advertise" behaviour, one would need to change the local binding behaviour to "only allocate (and advertise) a label for a remotely-learned IGP prefix x/y if you already received a remote LDP binding for this prefix or if you're the egress LSR for this FEC".. This is ordered control, something IOS only implements for cell-mode MPLS (i.e. ATM).

> But per node, you'd tell the nodes not to generate label, except
> for their loopback.

right, this part is simple..

> End result would be, that you'd only have loop0̈́s in each MPLS
> spakers LIBs, without any ACL/prefix-list maintenance overhead.

agreed. But I still see challenges getting this right in independent control mode.. Am I missing something?

 
>> But wouldn't a (prefix) ACL be enough to cover most cases? Generally,
>> loopbacks are allocated from one or more prefix ranges, so ACLs
>> could be rather static?
> 
> Yes, both can easily accomplish same goal, just bit additional admin
> overhead, while the true application in virtually all cases is, to generate
> label for single loopback interface. And actually we would have probably used
> 'your' way, had it been available when we wanted to implement it, instead of
> doing advertisement ACLs.

I guess so, filtering label allocation is more "natural" and efficient than filtering the advertisement for this very common case..

	oli
 


More information about the cisco-nsp mailing list