Hi Anton, I guess I see the use case described below as only one of the potential use cases for the X-AF tunnels. It seems that path computation, either head-end or PCE, could also use the dual-stack endpoint information. Note that the OSPF doesn’t establish the LSPs or even advertise the LSPs themselves– it merely populates the TE Database. I know that you know this but you want to assure your text doesn’t imply otherwise. Do you disagree? You can certainly keep this use case but I’d reference RFC 3906 (informational reference) and state that there could alternate use cases. Perhaps, your fellow author Alvaro (of OSPF TTZ fame) could help with some generic text preceding the specific IGP Shortcut use case.
Let me see if I can massage the backward compatibility text. I’m requested a Routing Directorate review and I’m going to start the LSR WG last call shortly. Thanks, Acee From: "Anton Smirnov (asmirnov)" <asmir...@cisco.com> Organization: Cisco Systems Date: Saturday, April 14, 2018 at 6:48 PM To: Acee Lindem <a...@cisco.com>, "draft-ietf-ospf-xaf...@ietf.org" <draft-ietf-ospf-xaf...@ietf.org> Cc: "lsr@ietf.org" <lsr@ietf.org> Subject: Re: OSPF Routing with Cross-Address Family MPLS Traffic Engineering Tunnels Hi Acee, sorry for my slow response. Before answering questions lets establish 'prerequisites' of the problem. - Network is dual stack, OSPFv2 is used to route IPv4, OSPFv3 is used to route IPv6 - TE LSAs are originated as per [RFC3630] and flooded in OSPFv2 - 'Endpoint' of each MPLS TE tunnel is IPv4 address - There is a desire to make OSPFv3 to compute IPv6 routes over TE tunnels - of which OSPFv3 has no topological information > 3. In the section 3 mapping algorithm, why do you walk the X-AF > endpoints from all connected areas? Why not just the area of local > IP address? Idea behind this wording is to cater for cases when area borders are laid differently in OSPFv2 and OSPFv3. It's even possible that router is ABR in OSPFv2 but not OSPFv3. From network design perspective this, of course, is a terrible thing to do - but not impossible. I guess I still don't understand. Are you implying that you are advertising TE LSAs using both OSPFv2 and OSPFv3 and aggregating the TED and since the area boundaries may be different, you need to search all the areas LSP endpoints? I don't think this deployment model makes sense and I don't think this should be supported. No, TE LSAs are advertised only in OSPFv2. Consider information available to OSPFv3 on tunnel headend router. Endpoint address of TE tunnel is IPv4 address, say 7.7.7.7 (this address is what tunnel tailend router advertises in OSPFv2 TE LSA in the Router Address TLV). OSPFv3 needs to find what router in what area corresponds to router that advertises that TE LSA in OSPFv2. That is, OSPFv3 has no its own TE information and not even a hint to which area may belong the tailend router. > 4. In the backward compatibility section, can you also discuss the > requirements for backward compatibility of the endpoints? Also state > that the X-AF tunnel will not be recognized unless the endpoints are > advertised by the same protocol (OSPFv2 or OSPFv3); or describe the > behavior if this is not the intension. We can add paragraph saying something like: "In order for XAF computation to work tunnel tailend routers MUST advertise XAF Node Local Address sub-TLVs in OSPF instance that will perform XAF computation. Thus only tunnel endpoints (both tunnel headend and tailend routers) and only OSPF protocol instance performing XAF routing must implement XAF as described in this document. Other routers in the network do not need to implement XAF algorithm or interpret Node Local Address sub-TLVs. For example, if network uses TE tunnels signaled by OSPFv2 [RFC3630] and intends to use cross-AF route computation in OSPFv3 then only OSPFv3 implementation on routers that serve as tunnel endpoints in OSPFv2 needs to be compliant with this specification." Will this text work? I think this could be a lot clearer if it were written from the perspective of the head-end router performing the calculation. Also, you lost me completely with the last sentence. We are uses a single protocol, OSPFv2 or OSPFv3 to advertise TE LSAs. Since both IPv4 and IPv6 traffic is tunneled over that LSP, there is no reason to operate both protocols since traffic will take the path of the X-AF LSP - correct? But OSPFv2 does not produce IPv6 routes. Both protocols operate in the network: - OSPFv2 computes IPv4 routes and distributes TE database - OSPFv3 computes IPv6 routes. If TE tunnels provide shortcut to destination then OSPFv3 will point route into the tunnel. --- Anton On 04/07/18 23:06, Acee Lindem (acee) wrote: Hi Anton, On 4/6/18, 7:33 AM, "Anton Smirnov (asmirnov)" <asmir...@cisco.com><mailto:asmir...@cisco.com> wrote: Hi Acee, my answers below (I didn't vet them with other authors, so they may express different opinions). > 1. Have you considered a shorter name for the RFC? For example: “OSPF > Cross Address Family Traffic Engineering Tunnels”? Your proposed variant drops two pieces: "Routing with" and "MPLS". Dropping mention to MPLS is fine with me. Dropping "Routing with" seems to me less correct because the draft is about ways to compute routes and not about setting up/managing tunnels. But ultimately I have no strong feelings here and if there is a requirement to shorten document's name then that would be a good candidate. > 2. Can you change the requirements language text to the RFC 8174 version? OK, we will publish new document revision when we agreed on other points. > 3. In the section 3 mapping algorithm, why do you walk the X-AF > endpoints from all connected areas? Why not just the area of local > IP address? Idea behind this wording is to cater for cases when area borders are laid differently in OSPFv2 and OSPFv3. It's even possible that router is ABR in OSPFv2 but not OSPFv3. From network design perspective this, of course, is a terrible thing to do - but not impossible. I guess I still don't understand. Are you implying that you are advertising TE LSAs using both OSPFv2 and OSPFv3 and aggregating the TED and since the area boundaries may be different, you need to search all the areas LSP endpoints? I don't think this deployment model makes sense and I don't think this should be supported. > 4. In the backward compatibility section, can you also discuss the > requirements for backward compatibility of the endpoints? Also state > that the X-AF tunnel will not be recognized unless the endpoints are > advertised by the same protocol (OSPFv2 or OSPFv3); or describe the > behavior if this is not the intension. We can add paragraph saying something like: "In order for XAF computation to work tunnel tailend routers MUST advertise XAF Node Local Address sub-TLVs in OSPF instance that will perform XAF computation. Thus only tunnel endpoints (both tunnel headend and tailend routers) and only OSPF protocol instance performing XAF routing must implement XAF as described in this document. Other routers in the network do not need to implement XAF algorithm or interpret Node Local Address sub-TLVs. For example, if network uses TE tunnels signaled by OSPFv2 [RFC3630] and intends to use cross-AF route computation in OSPFv3 then only OSPFv3 implementation on routers that serve as tunnel endpoints in OSPFv2 needs to be compliant with this specification." Will this text work? I think this could be a lot clearer if it were written from the perspective of the head-end router performing the calculation. Also, you lost me completely with the last sentence. We are uses a single protocol, OSPFv2 or OSPFv3 to advertise TE LSAs. Since both IPv4 and IPv6 traffic is tunneled over that LSP, there is no reason to operate both protocols since traffic will take the path of the X-AF LSP - correct? Thanks, Acee --- Anton On 04/04/18 20:13, Acee Lindem (acee) wrote: > Hi Anton, Alvaro, and Mike, > > In preparation for WG last call, I have a couple comments. > > 1. Have you considered a shorter name for the RFC? For example: “OSPF > Cross Address Family Traffic Engineering Tunnels”? > 2. Can you change the requirements language text to the RFC 8174 version? > 3. In the section 3 mapping algorithm, why do you walk the X-AF > endpoints from all connected areas? Why not just the area of local > IP address? > 4. In the backward compatibility section, can you also discuss the > requirements for backward compatibility of the endpoints? Also state > that the X-AF tunnel will not be recognized unless the endpoints are > advertised by the same protocol (OSPFv2 or OSPFv3); or describe the > behavior if this is not the intension. > > Thanks, > > Acee >
_______________________________________________ Lsr mailing list Lsr@ietf.org https://www.ietf.org/mailman/listinfo/lsr