Rob - I haven't seen anyone on this thread suggest that if a router ID is available/reachable that it should not be preferred/used. Local policies could make any particular address unusable, but it would seem unusual for a customer to configure a box to advertise a /32 which it also prevents from being used. So I think Peter's point is that it is not unreasonable to consider the set of all reachable /32 which a node has advertised as being viable candidates.
Frankly, I find the discussion of a preference algorithm in selecting the endpoint address as useful/interesting - but much more appropriate for a vendor deployment guide than a normative specification. Vendors often are faced with idiosyncratic deployment constraints from their customers which need to be accommodated. In which case responsive vendors will provide various knobs to allow override of default behavior - while retaining the ease of "zero config" for the majority of customers. This is simply good business. We should not attempt to "standardize" this. Les > -----Original Message----- > From: [email protected] [mailto:[email protected]] On Behalf Of Rob > Shakir > Sent: Thursday, December 13, 2012 11:56 AM > To: Peter Psenak (ppsenak) > Cc: [email protected]; Stewart Bryant (stbryant) > Subject: Re: identifying IP address of targeted LDP session in draft-ietf- > rtgwg-remote-lfa-00 > > Hi Peter, > > On 13 Dec 2012, at 14:25, Peter Psenak <[email protected]> wrote: > > > > local algorithm that picks any of the /32 IP addresses advertised by PQ > node will work in 100% of cases. > > If there are local filters or control-plane protection ACLs deployed on the > target node to which the T-LDP session is to be established, it is possible > that these drop T-LDP traffic not targeted to a particular local address. I'm > not clear on how this can be determined remotely? > > It seems to me that we need deterministic behaviour in order to allow > operators to clearly consider this within such policies (or to know that > other protocol/signalling mechanisms are required), and to ensure inter- > operability. I don't see a clear reason why Hannes' suggestions do not go at > least some way to providing this. In the vast majority of deployments I have > seen, a consistent router ID is used for intra-domain, inter-node session > termination and protocol identifier. This is more common (imho) than the > highest /32 being the most suitable address. > > r. > _______________________________________________ > rtgwg mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/rtgwg _______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
