RTG WG IETF 91 (Hawaii) Minutes WEDNESDAY, November 12, 2014 900-1130 Morning Session I Coral 2 ===================================================== CHAIR(s): Alvaro Retana Jeff Tantsura SCRIBE: Acee Lindem - Administration - Alvaro Retana - Charter Discussion - Alvaro Retena (See slides) * Multicast Architecture presentation moved to PIM session. PIM will re-charter to cover complete range of multicast topics. * No comments at meeting on RTG WG charter and NWP (New Work Process) * Send commments to the list. - Use of BGP for Routing in Large Scale Data Centers - Jon Mitchell (See slides) Jabber Question From Adrian Farrell: How does this relate to work in BESS? Jon: Unrelated to any work in BESS. Lou Berger: Adrian asking about L3VPN or L2VPN? Jon: No VPNs or MPLS in design now. Could be added. Jeff Tantsura: At least one other MSDC is extending MPLS into data center. Lou Berger: Is BGP used to set up VPNs. Jon: No - VPN provisioning is controller based. Sam Aldrin: Any tenant provisioning with BGP? Jon: No. Acee: Where is the aggregation amongst Tier 1, Tier 2, and Tier 3? Jon: Minimal policy. Routes go everywhere other than down to servers. Will add more information to draft. Alia Atlas: Can you address comments from Sue Hares? Jon: I believe I have done this and will follow up. Jeff: How is remove private AS used? What is the draft? Jon: Needed to avoid influencing routing in core. Jeff: Do you require sub-100 millisecond convergence? Jon: Don't have requirement today. Some failure already do meet this requirement. Jeff: Expand a bit on modeling failure types. Alvaro: RTG WG needs to provide more review. Please read and comment. - Extended procedures and considerations for evaluating Loop-Free Alternates Uma Chunduri (See slides) Acee: How would you do the inheritance of multi-homed prefixes any other way? Uma: Could do a general per-prefix computation. Acee: Why would use a link for backup that has an IGP maximum metric? Uma: We are not relaxing the forward direction checking - just the reverse path IGP maximum metric check. Stewart Bryant: Multiply attached prefixes was addressed in earlier tunneling draft. - LFA selection for Multi-Homed Prefixes - Pushpasis Sarkar (See slides) Uma: Helpful for ISIS. Acee: Why is there a problem of backward compatibility since this local repair? Pushpasis: It is not OSPF LFA compatbility, it is RFC 1583 compatibiltiy. Alia Atlas (No hats): This work may be necessary now that there is more implementation experience. Alia: We may be able to open the RFC 5286 as a BIS. If not, this clarifying draft could be useful. Alvaro: Any thought of taking on the BIS work? Alvaro: Take it to the list. - A YANG Data Model for Routing Management - Ladislav Lhotka (See slides) Rob Shakir: Routing Instance is VRF? Lada: Yes - that is one type. But there are other use cases. Rob: Doesn't like different vendor model implementations. Lada: Takes too long for standard model. Rob: We don't want vendor specific models Jeff Haas: 2 things pulling us in different direction, unclear what is routing instance, wrt routing, management, etc, L3VPN - don't see model where a route could be received by a protocol and then distributed to another protocols. Lada: It is simple - every RIB belongs to its VRF. Jeff H: More indicies, less contents. Acee: There's a problem with route-filters, should be in a separate routing policy model rather than stubs in this basic model. Lada: It is a basic model, it defines the points where you could hook route-filters Acee:Just remove the route filers. The fact that you don't take comments is what is causing this draft to take forever. Acee: For interfaces, you should augument IP address with a routing instance reference, you can't change an existing RFC to match your model. Lada: I don't know why we didn't do better - perhaps becasue you didn't comment earlier. Alia: we work together to get coherent model working for routing, people like Acee with lots of real world expirience provide rather important comments. Dean Bogdanovic: Where would you put the route policy? Acee: Separately, it is not useful to keep packet classification and route filtering together. Derek Yeung: Logical routers, we have interdependancies wrt interfaces as per existing implemtations. Jeff Zhang: Filtering can be between protocols in the same RIB. Lada: Model can handle this. Dean: Augment with separate policy model. Rob: Sees filters as useful. Not sure exactly where they belong although is also policy in the proposed BGP model. Xufeng Liu: Would like concensus on model. Alvaro: Netmod will also discuss this draft tomorrow at 4:00 PM. - Destination/Source Routing - David Lamparter (See slides) Fred Baker: Would multicast not be related to SSM? David: ISPs can handle this in the home net. Doug Montgomery: Can you do strict RFP checks at egress points with this approach? David: This doesn't change the picture. This insures the home net will conform to egress filtering. Doug: Is this safe now. David: Yes Adrian: Are we jumping through loops to conform to egress filtering BCP? Are there other simpler mechanisms. David: I don't see this as being hard to implement. Adrian: Not hard to implement but hard to deploy in a legacy network. David: I would agree it is not easy. Targeted for home net where this isn't an issue. Fred: Will be taking to implementer of OSPF this week. Alvaro: There are other use cases for source/destination routing. Rob: Hetrogeneous rules for screening subscribers would be untenable. Stewart: We should find the simpliest solution. Maybe we could get providers to filter based on multi-homing for a fee. David: Believes this is a better solution and that it not that difficult to implement. Rob: Two sides to debate. Is this a problem? What is the solution? Anytime you need to get tech support involved, it is very expensive for large ISPs. Fred: What are the complexities that are involved with mixed deployment? Routing protocol is not that more complex. FIB lookup and addition are also changed. Alvaro: Is there a problem? People in room all think there is a problem. We will be discussing this on the list. - A Framework for Secure Routing Protocols - Bill Atwood (See slides) Acee: Key-chain YANG models are under development. Bill: I was not aware of this. I will talk to you. Adrian: Need proper draft name on this. Ilich (sp): How does this work? What is the recovery? Bill: Imagines how it might work. Uma Chunduri: KARP had no interest in KMP. These heavy databases will not fbe needed. Acee: This framework offers no benefit over a key-chain YANG model.