PPVPN WG meeting at 55th IETF, Atlanta, GA - Nov 20, 2002 Minutes recorded by Ananth Nagarajan (few additions/changes by Marco Carugi) Marco started with introducing the agenda. Meeting related documents will be posted on Nortel-maintained web site. 1. WG status ------------ Milestones have been recently updated. Layer 3 reqts and framework have been submitted to IESG. We are now at second round of comments, before they move towards Informational RFC. L3 approaches and applicability statements : submission to IESG for publication will be hopefully started in December. Marco also described the status of various documents (these updates have been sent to the mailing list). Last call on draft-ietf-ppvpn-cl-tunneling-vpn-00.txt not successful. Solicited input on whether this should be progressed in the WG. Little support, but most didn't care. Scott - Lack of strong opposition does not mean support. Under WG consideration - generic requirements, l2vpn requirements, vpn DNS-based discovery documents, terminology and metrics. Kireeti - Move to WG - does it mean that it is moved to WG status? If so, it needs call for consensus. Marco - yes, the normal process will be followed. Scott - CL tunneling: Is there support for publishing this as a informational RFC, as opposed to BCP? ASk for consensus on BCP RFC, Info RFC or is against publication? Yakov - BCP means best current practice. Since it is neither, there should be no such option. Scott - Disagreed. Consensus - support from authors for info RFC. None for BCP. About 10 people against publication. A lot of agnostics. Decision - need more discussion on mailing list. Scott - IESG has enough info to take this off the WG agenda. WG last call for draft-nagarajan-ppvpn-generic-reqts after it becomes a WG ID, to be done by end of 2002. Conditional to IESG approval of L3 reqts and framework, 2547bis and VR approaches to be advanced to Proposed standard. Protocol dependencies to be solved and submitted in parallel, together with applicability statements. Other 2547-based solution documents, VR extensions, L3 solution-specific MIBs etc will be advanced in reasonable timing. CE-based IPsec solution has not been updated since Yokohama, along with corresponding AS. Would like to progress this quickly. Design teams: Requirements team task on generic requirements and improvement of L3 and L2 requirements is being done. Some issues: Cooperation terms with IEEE 802.1 need to be figured out. Interaction with related WGs in IETF needs to be done for PPVPN, independently from whether sub-ip area survives or not. Layer 2 interworking for like-to-like is ok, while any-to-any is not currently ok (out of scope), apart from possible specific scenarios. Other work to be progressed includes QoS, management, security framework, multicast etc. for PPVPNs. Input from community is solicited. Liaison received from ITU-T on L1 VPNs was discussed. 2. Generic Requirements - Ananth Nagarajan ------------------------------------------- Ananth described the generic requirements draft and solicited input, especially for the scalability section. Loa talked about the terminology draft that is being simultaneously developed. There was a word of caution on the way this terminology draft would be progressed separately, as it would create problems for proposed solutions if they don't comply wih this. Request acceptance of generic requirements as WG draft. General consensus was for this acceptance. 3. CE-CE Authentication - Ron Bonica ----------------------------------- Ron made a short presentation on main issues related to this work. Process considerations - need BGP extended community type. Need UDP port number for new protocol. Vach expressed a procedural concern about whether PPVPN can do protocols. 4. CE autoconfiguration for CE-based PPVPN - Cheng Yin Lee ---------------------------------------------------------- Cheng Yin described the draft. Issues - is being able to push/update necessary? Is polling sufficient? Combining polling with receive-trigger-then-pull or push approach may be a good compromise (may then reduce polling interval)? 5. IPsec protected Virtual Links for PPVPNs - Mark Duffy -------------------------------------------------------- Virtual Link - point to point link across IP network implemented via tunneling. This is envisioned for CE-based and PE-based vR solutions. This can also be done between CE and PE. It is also useful for 2547, and hybrid networks for securing links between PEs running 2547 and VR. Requirements - multi-context support (multiple virtual links between a given pair of systems). Eric - Does it make sense to have one security association for all vpns, as opposed to separate associations for different vpns. Mark - Yes. Why this is needed? Current efforts don't address the multi context needs of network based VPNs. Issues : How does IPsec policy fit where forwarding decisions are made by best-matched prefix? How should VPN-ID like context be conveyed when a tunnel is established. Draft describes various approaches using combinations of either IPsec tunnel mode or IPsec transport mode with other tunneling schemes. Applicability to ppvpn - standards-based virtual links for adequate security for interoperability between approaches. Eric - There is already a draft on IPsec with 2547. Why is this different? Mark - this is not objecting to that draft, but that is specific to 2547 context. This is more generic. Marco - Is this mandatory before progressing VR? Or is it complementary? Mark - There is no mechanism defined today for securing VR tunnels. If you simply say "use IPsec tunneling between VRs", everyone will be doing it in a different way. Marco - Need further discussion on this. Mark - There is no alternative available in current drafts. Yu-cheng Ren (ISI) - This topic has been addressed by various drafts. Why do we need another (e.g. draft-touch, draft-knight). Mark - these drafts talk about using ipsec tunnels, but doesn't talk about securing multiple links. One of the authors of draft-touch - this is not the case. It talks about ip-in-ip draft. Yu-cheng - Nothing against this draft. Can this not be integrated into framework or requirements drafts? Mark - some part of this can be added. Need discussion on what needs to be included. 6. Use of multiple instance OSPF for PE-CE - Kunihiro Ishiguro -------------------------------------------------------------- Kunihiro presented this draft - "simple way to use ospf for ce-pe". Procedure - Assign OSPF instance to VRF. Benefit - much simpler, can coexist with ospf configurations, and can be applied to is-is. This draft complements Rosen draft (draft-rosen-vpns-ospf-bgp-mpls-05.txt) and overlaps a little bit. Eric - Don't need different procedures to do the same thing. Hoped this would eliminate complicated things in existing mechanisms. 7. L2 VPN design team report - Loa Andersson and others --------------------------------------------- Documents - metrics, terminology, framework (not much work done on these three). L2vpn requirements - significant progress. There was a design team debate on emulated LANs versus bridges. The design team has 22 members, plus several others who contributed that are not on the team. Recommend that design team structure be discontinued and work be continued separately. Requirements - Yetik Serbest ------------ Status - evolving towards common l2vpn requirements (as opposed to individual requirements for vpws, vpls etc.) Aligned with generic requirements draft, definitions moved to terminology draft, sections added on referencing generic reqts, addressing, network resource partitioning/sharing, access control, interoperability. Future work - clarify any issues with what a vpls is, and what it does, and clarify vlans in vpls. Also request feedback. Request moving it to WG status. Loa: The document has a "point-to-multipoint" VPWS, which is wrong and needs to be removed. A: Agreed. Cheng Yin: Is VPLS emulating a bridge or LAN? Yetik - wait for Eric's presentation. Architectural Model - Eric Rosen -------------------------------- Eric described the architectural model, and addressed points of disagreement between IETF and IEEE people, and inter-relationships between VPLS and IEEE bridge/LAN emulation. Other issues - Spanning tree, presumed LAN properties including packet order preservation, reliability, non-duplication, low latency. Loa - Nice powerpointing picture will be converted to ASCII art and used in requirements and framework drafts. Issues of VPLS - Muneyoshi Suzuki --------------------------------- Muneyoshi discussed issues related to customer views, provider views, vpls support etc, routing for LAN/bridge emulation. Norm Finn - IEEE 802.1 unofficial liaison ----------------------------------------- Project authorization request (amendment to 802.1q VLANs) - enable a SP to offer the equivalent of separate LAN segments, bridged or virtual bridged LANs (network of bridges doing l2vpn type services). Name of project - 802.1AD. MAC-in-MAC not possible with this amendment. Confident that IETF L2vpn and 802.1AD will be interoperable, compatible etc if mailing discussions are carried through. Q: how to solve MTU in 802.3 problem? A: Make it longer. This is possible. Cheng Yin - There should be minimum of overlap with vpls and IEEE work. But there will be overlap as IEEE defines Q-in-Q and IETF defines other encapsulation. A: Outer Q tag can be used as Pseudowire header. So there won't be a problem. This is exactly VPLS - just a different way of looking at it. Marco - intention from both sides to minimize overlap. Vach - there is no standard q-in-q way to identify customer packets. What the IEEE is doing in standardizing this may be a good thing. 8. CE-based Virtual PRivate LAN - Cheng Yin LEe ---------------------------------------------- Previous version described use of l2tpv3 for this application. l2tpv3 signaling and tunneling will be taken to l2tpext WG. This model describes the transport of Ethernet over IP tunnels (CE-based). Future work includes failure scenarios, monitoring of vpl connectivity and other l2tpv3 specifications to be done in l2tpext WG. Eric Rosen - Why not IPsec as it is CE-based? Why do use l2tpv3? Cheng Yin - No protocol number for use of ipsec. 9. Identify PW endpoints signaling - Eric Rosen ----------------------------------------------- It is possible that PWE3 owns the signaling portion. l2vpn can be built over pwe3 pseudowires. Problem statement: L2vpn has various autodiscovery. Martini signaling presupposed provisioning models. This solution identifies ways to use martini signaling and make it more applicable to l2vpn. Forwarders will be identified by vpn-id. Eric described various techniques (e.g. Distributed VPLS, VPWS full mesh, switched connection i.e., ATM SVCs to pseudowires). RAhul Aggarwal - Support this as it solves real issues with martini approaches. Marco - need clarification from ADs about this space, and where this needs to be done. Eric - Some of these also apply to l2tp signaling, in addition to martini Yakov - Can you use the form of identification used by the Route Target BGP Extended Community Eric - yes. Hamid - suggested to use global unique identifiers instead of VPN-IDs Marco - Global ID issues need to be taken to list. 10. Scalability issues with VPLS - Andrew Sodder ------------------------------------------------ This draft has various parts, one part belonging to ieee (mac in mac), other parts to ietf. The draft also proposes vpn identifier and control word use. Many issues left to be addressed - vpls interoperabilty, oam ID work to be done in IEEE. Request continued discussion on list. Rahul Aggarwal - Don't understand motivation for the draft. There is no scaling issue with current mechanisms. This draft request changes to wire format, and requires changes to hardware. A: Depends on who you talk to. Rahul: Depends on which specific vendor box you are talking about. 11. GVPLS/LPE - Dinesh Mohan ----------------------------- Goal - support seamless integration of both distributed and non-distributed vpls models. Allow integration of different access topologies, in a hierarchical way. Other objectives - scalability and performance, optimizaton of replication and support for multicast applications, interoperability with other vpls solutions, integrated oam capability in data plane. The complementary draft-chen-ppvpn-dvpls-compare-01.txt describes comparisons of vpls approaches. Andersson's metrics draft was also mentioned. Future steps - convergence with solutions drafts, signaling, QoS/resiliency, enhance oam capabilities. Marco - comparison draft and metrics draft should be used by the design team. 12. VPLS based on IP multicast - Ali Sajassi ---------------------------------------------- Multicast, if supported on backbone, can be leveraged for vpls. mp2p PW for unicast and mp2mp PW for multicast is proposed. Encapsulation is GRE or l2tpv3. pw signaling is not needed. Issues - packet reordering may occur. Inter AS and IP address management. Sasha Vainshtein - PPVPN scalability parameters in requirements draft not enough for feasability of this draft. Also, all routers in provider core would need to learn routes to all these addresses. Text of the draft doesn't say how unicast addresses are treated by the device (pingable, or something else). A: RFC 1918 specifies private address space. This scheme can be used in hierarchical way (access, core etc.) and thus may tackle scaling issues. Regarding router issues, these can be taken to the list. Eric - VPLS depending on IP multicast sounds foolish. learning based on source address is also not a good idea (fragile). A: Discussion on list Juha - One of the requirements is operability in multi provider environment. Multicast may not be stable in this scenario. We need to take that requirement into account. 13. L2VPN Interworking - Ali Sajassi ------------------------------------- Local attachment circuit termination proposed for various attachment circuit types including ATM, ip, ppp, multiprotocol. Need two new PW types for IP and multiprotocol. Rahul - It may be useful to consider the idea of a "raw" pseudowire that can be used for multiprotocol case. Ali - wouldn't apply for multiprotocol, as it needs identification of protocol. Scott - Interworking out of scope of, when sub-ip was set up. Ali - purpose is not to reinvent the wheel. Just suggesting what pseudowire to use. Marco - Scott, should this continue to be discussed in the Working group. Scott - if it is restated as it was by Ali, it can be done. 14. Summary - Marco ------------------- In Yokohama, it was agreed to propose solutions based on functional structure proposed in L2 solutions design team. Next work steps on L2 were discussed. Functional decomposition of l2 is strongly recommended. No protocol development in ppvpn. Limited set of solutions, so it is not recommended to progress "all theoretically possible solutions". So we need to determine a pragmatic set of alternatives. Main functional blocks identified by DT (various options per block) - discovery, signaling, pw tunneling/data plane forwarding, informational model. Therefore, the functional documents should have a requirements section, option specification section(s), a section on interactions with other functions. Possible organization of deliverables : - N documents per each functional block, one per option - one doc per each functional block (N option-specific sections). A number of mono-option documents is almost done. Different solutions using same option for a specific function should rely on the same functional document. Aggressive plan for L2 DT is to have a DT meeting in January 2003 (1-2 days). Tasks to be completed were identified. It is possible to have another design team (in addition to this one) before the next IETF meeting.