Hi Julien, I really like the discussion. It has already produced some very good results.
Please, see in-line. Igor ----- Original Message ----- From: "MEURIC Julien RD-CORE-LAN" <[EMAIL PROTECTED]> Hi all. Allow me to get back to the issue raised during the L1VPN meeting: the autodiscovery protocol. I think it is clear for everyone that BGP really fits the job. Thus I do not see why we would need to add an IGP to do less things. Lots of drawbacks were even pinpointed during the meeting: more flooding, less scalability, no AS crossing, less flexibility (full mesh)... but no real advantage. IB>> Just a couple. 1) No need for BGP. Untill now operators of L1 layer networks don't seem to find any use for BGP, and unlikely to deploy BGP for the purpose of one application some time soon [JM] OK, but BGP is already deployed for this over some networks other than L1. Autodiscovery is not L1 specific and I do not see the need to introduce too specific mecanims when it is not necessary to be so specific. As soon as we have started to deploy a whole IETF artillery (IP + LMP + OSPF/IS-IS + RSVP-TE + ...) to introduce a CP into transport networks, [IB] We are not talking about whole IETF artillery here. We are talking about tiny OSPF extensions, very similar to what are currently working in any GMPLS controlled network I feel that we should use already well known bricks too build the wall, and avoid to fill gaps with (ad-hoc) half bricks. Moreover, I believe that multi-AS is a key feature to support. [IB] I really would like to have a multi-AS L1 network operator as a customer of Movaz :=) However, I am yet to see a multi-area L1 network. You have mentioned interop issues between the two solutions and inconvenience supporting them both, and I am afraid you are missing a point here. In IP network BGP and OSPF work together just fine - they both populate the same RIB. How is it different from populating the PIT, which what L1VPN auto-discovery is all about? It certainly could be populated from many sources - BGP, OSPF, configuration, etc. I don't see any issues for working the two mechanisms simultaneously in the same (say, multi-AS) network. Would you disagree with that? 2) IGP solution facilitates easier integration with TE, which is very important for GMPLS based control plane [JM] I must concede this. What is more, as Kireeti said, starting with 2 different solutions is likely to bring more problems than solve any. Since something needs to be added, I am not sure that extending an IGP (or both if we do not want to preclude any option...) will be easier/quicker than using BGP. IB>> As was pointed out on the meeting, this is not an argument of a sort RSVP-TE vs CR-LDP. The architecture (basic mode) clearly assumes the protocol independence of L1VPN auto-discovey mechanism. The OSPF extensions are very simple, the application is required now and based on OSPF could be delivered quickly. What happens if the BGP solution proves to be a better one (as it happened in the past) ? Than the OSPF solution could be simply thrown away and replaced by the BGP one. [JM] I agree with the scenario, but this introduces another issue which is migration, and that is not the easiest part. [IB] Please, see my comment above Furthermore, we are not in the position where we think about extending IGPs until a new solutions shows up: BGP already exsits, and the draft was ready even before. [IB] Well, the OSPF extensions would probably take 1 month to implement (including extensive testing) and, as was said before, even basic BGP is not deployed on L1 network Two more points: 1) Scalability concerns. True, Ps do not need to have anything to do with VPN information, and in OSPF solution they still need to participate in flooding of L1VPN LSAs. However, OSPF is designed to distribute and manage large numbers of external routes (AS-scope type LSAs). We do not need external routes in L1 network - that's why we do not use BGP in L1 network in the first place. Distribution of AS-scope L1VPN LSAs carrying VPN info seems to be far smaller and far less dynamic substitution for type 5 LSAs 2) If IGP based discovery is such a bad thing, why don't we do discovery of any type via BGP? For instance, there are reasons why we are doing, say, PCE capabilities discovery via IGP. Why L1VPN discovery is any different ? Igor _______________________________________________ L1vpn mailing list L1vpn@lists.ietf.org https://www1.ietf.org/mailman/listinfo/l1vpn
