********************************************************************** IETF 83 PWE3 - THURSDAY, March 29, 2012 - 09:00-11:30 (150 min) Room 242AB ********************************************************************** Chairs: Matthew Bocci and Andy Malis Secretary: David Sinicrope (X slide sets NOT received as of Mar 29 00:43 Paris time) 20 min - Agenda bash, WG Agenda and Status - Andy MALIS and Matthew BOCCI Matthew presented the WG status slides. See slides for details. Goals and Milestones need to be updated. Mattew updated what was discussed at last IETF. Some of the dates have been updated to be more realistic. Please post comments on the goals and milestones to the list. static-pw-status - Stewart: needs author signoff from Luca and need to determine whether it needs a short last call or can go out. packet-pw - a couple of IESG comments need addressing. A respin will be done after IETF83. pwe3-mpls-eth-oam-iwk - new version on the way, but missed deadline. Will be posted after IETF 83 p2MP-pw-requirements - Stewart mix of requirements and architecture. From architecture framework point of view, it lacks clarity of other architecture documents. Necessary to have clarity because it gives guidelines for solutions. Need a short but clean approach that incorporates SS and MS case. Chairs looking for someone to write the document because it will hold up the rest of p2MP work. THe Chairs urged the WG to step up and address the document to not hold up the work. Also need to cover the OAM for P2MP, or shouldn't do P2MP. The Chairs noted a number of drafts they felt were ready for last call and asked for any comments from the room. It was noted that ICCP should relate closer to PW redundancy. 10 min - A Unified Control Channel for Pseudowires - Tom NADEAU http://tools.ietf.org/html/draft-ietf-pwe3-vccv-for-gal-00.txt (No Slides) Handled as part of larger discussion 20 min - General discussion on VCCV for GAL There appear to be two separate issues on the table: 1. VCCV support for a PW associated channel that uses the GAL as an alert mechanisms (VCCV Type 4). This is really the subject of the draft. 2. What to do about legacy modes, particularly router alert and TTL expiry. This is a separate issue regarding the progress of RFC5085 (or some future RFC) to Internet Standard status. New draft was built on feedback from Nick's survey. The goal is to simplify the modes for interop and deployment. VCCV use was simplified down to a mode with and a mode without the control word. Outstanding question: What to do with the other modes? How do we make them optional with out impacting existing implementations. Two optons overall: 1. dont' address in this document. Addressed in 580bis. This document only defines new GAL type. Is this structure OK with the WG? Proposal is to deal with the removal of some of the old modes in RFC 5085bis rather than the VCCV draft? General support for dealing with this in a 5085bis. 2. How do we handled the old modes? capability advertisements, how do you deal with that and bring it forward? YJS: Can remove for architectural reasons. For TTL expiry, how do you deal with a remote node that doesn't deal with the new mode or capability advertisement. Tom: could do capability negotiation in additon to advertisement YJS: Need to separate CP from data plane. Not everyone is going to change what's out there. Much out there that is not TP based. Not going to change anytime soon. Carlos: right approach to create a 5085bis, separate data from control plane. Can get creative with what does it mean to have a CC type at all. CW is sent with TTL 1 regardless. What does it mean to have a TTL expiry with or without the CW. Nick: not going to solve this here. Having a bis will generate discussion. The survey is a snapshot not everyone responded. The use of CW vs TTL was roughly even. Worst case is 4 flavors, or can we reduce the number of cases. Carlos: If we send VCCV Type 1 with a CW existing implementations don't complain. We may be able to simplify the CP. Tom: could do a fence around old modes, but what is burden on implementations. The goal was to simplify down to two modes one most popular and another to support non-CW case. YJS: how does this simplify things? In words it sounds like 2 modes is better than 4 modes, but to remove modes you make the CP more complex with advertisement and negotiation. How is this simplifying. Greg Mirsky: If we have depricated TTL as an explicit VCCV Type, does it mean we don't have to react to TTL expiriation? Tom: you could ignore it. Greg: but creates ambiguity and inconsistent behavior. YJS: also inconsistent with IP use of TTL. Tom: can silently drop router alert Greg: without TTL you can't address SP, should leave it. Tom: then need to make a conscious decision to keep it. YJS: simplest thing to do is not to pair things down now. In 5 years if folks are not using old modes then simplify. Nick: tend to agree, but issue is still interoperability. With 3 modes having issues with different camps. With 4 makes more difficult. YJS: do another survey in 5? years to see if anyone using TTL. Nick: Could do some with standards by saying you MUST do Dave Sinicrope: BBF has depricated the use of Router Alert in their MPLS in Mobile Backhaul and MPLS in Carrier Ethernet implementation agreements using VCCV. Dave Allan: Also in their MultiService Architecture Himanshu: can't depricate TTL Matthew: should we leave in router alert in VCCV? Matthew: Should we remove router alert from VCCV? <~ 20hands> Tom: should we start rolling out 5085bis? Matthew: Yes YJS: Do you need a co-author on 5085bis? Tom: Yes Andy: recorde that 5085bis has a co-author (i.e., YJS) Will take the issue and rest of the discussion to the list 10 min - VCCV MPLS-TP Connectivity Verification (CV) Capability Advertisement - Greg MIRSKY http://tools.ietf.org/html/draft-mirsky-mpls-tp-cv-adv-00 Greg Mirsky presented proposes a method of advertising capabilities. Carlos: is this mandatory? Greg: not mandatory, Carlos: are we going to run into the same problem down the road because we are using a bit mask? Greg: no, there are reserved bits for extension. YJS: the VC and the CV are not quite right. Can we call it PW CC? Greg: could do this via TLV. If there is a strong opinion we could do it via TLV. Carlos: creating a sub TLV and already using 1/2 Greg: no, only using 25% Carlos: to make future proof might want to look at TLV Matthew: MPLS-TP OAM could potentially be used for PW ???: Greg: could be used operationally, this is a copability advertisement. At the end you would expect to come to an operable state. Discussion will be taken up on the list. 10 min - Static pseudowire configuration checking using Generic Associated Channel (G-ACh) Advertisement Protocol - Lizhong JIN http://tools.ietf.org/html/draft-jc-pwe3-mpls-tp-static-checking-00 Lizhong presented Note the acronym GAP. Not the clothing store. This is a draft in the MPLS WG as well. Loa: There are two drafts related to this in MPLS, a generic and a -tp draft. If you want to make it generic, you need to remove the TP acronym Dan Frost: If you want to understand this you need to look at the generic draft not the -tp draft. YJS: Did we require in the original PW, that CW be used in both directions? In static you can configure it differently. In a static PW what does FEC 129 mean? Lizhong: for MPLSTP we wanted to only use FEC 129. Matthew: FEC 129 would be for signaled PWs, but for MPLS-TP with static PW, you want to use FEC129 like parameters. No FEC 129 Sri: Problem is bigger than PW. Lizhong: This feature will depend on development of GAP protocol. This is only done for PW. xxx IP Infusion: what is frequency of GAP messages? Lizhong: suggested times are in the draft Matthew: as GAP co-author, advertising things at the segment layer. When you use at PW layer you need to be careful about timings because of scale. Stewart: Should CW be identical? RFC 4447 - the LDP version doesn't use CW in the event of a mismatch. YJS: only if you use LDP, static no way to enforce. Architcture doesn't say so each end could be different. Stewart: not only provising different, but asymmetrically. Need someone to research discussions and determine what was agreed. YJS: I'll take action to research and write it down perhaps in RFC5085 bis. Discussion will be taken up on the list. 10 min - LDP Extensions for Lock Instruct and Loopback of Pseudowire in MPLS Transport Profile - Jie DONG (presented by Greg Mirsky) http://tools.ietf.org/html/draft-dong-pwe3-mpls-tp-li-lb-00 Greg presented slides Proposal uses two new flags out of the OAM Admin TLV To exit states, the bits are cleared Samy: when writing 6435 we agreed to not have two mechanism a control and an OAM mechanism. Greg: because OAM is inband, the processing of OAM messages when already locked is a special case. In control plane you can do this. YJS: This is a standard procedure and we've done it many times and you just watch for the packets. Greg: it was proposed for the document but taken out. This matching proposal from CCAMP is to have the same mechanism in the control protocol Himanshu: agree with Samy that it should be part of OAM protocol YJS: simple question, understand there might be a security problem if loss of control plane, are we stuck in Loop forever? mumbles: no Discussion will be taken up on the list. 10 min - Signaling Root-Initiated Point-to-Multipoint Pseudowires using LDP - Sami BOUTROS http://tools.ietf.org/html/draft-ietf-pwe3-p2mp-pw-04 Samy presented the slides Samy: The draft contains an implementable solution and is ready for last call, but there is an issue with the requirements draft. Andy: Yes, this draft won't be last called until we have a resolution on the requirements draft. 10 min - P2MP PW protection for mpls-tp network - Guoman LIU http://tools.ietf.org/html/draft-liu-pwe3-mpls-tp-p2mp-pw-protection-00 Guoman presented the slides. YJS: Saying each of the CEs dualed homed to the network. WHy do you need node E? No connectivity to A. Guoman: YJS: assuming no control plane in the backwards direction Matthew: are you assuming this is a ring Guoman: no not assuming a ring. Matthew: for MPLS-TP and deals with P2MP. P2MP for MPLS-TP hasn't been progressed yet. Should see where P2MP MPLS-TP architecture is going in MPLS first. Loa: If you have P2MP LSPs, then PW is pretty much solved, then source to receiver only need to signal to the end-node Matthew: this was protection for P2MP PWs. Loa: that's way out of scope for anything we have now. Need to see what basis we have to build on for this. We sometimes have drafts that are out of scope, but ok to publish them and need to keep them somewhere for when it is time. Matthew: need to progress the architecture for P2MP first 10 min - LDP Extensions for Pseudo Wire (PW) Transfer in an MPLS-TP Network - Qilei WANG http://tools.ietf.org/html/draft-bao-pwe3-pw-transfer-00 Send email to the list requesting comments on the draft. 10 min - LDP extensions for Explicit Pseudowire to transport LSP mapping - Mach CHEN http://tools.ietf.org/html/draft-cao-pwe3-mpls-tp-pw-over-bidir-lsp-05 Ping Pan presented the slides requesting WG draft Himanshu: good draft, missing functionality it addresses. In PW we have libral retention, unsolicited, may not have a PW with a PSN binding. Do you have steps to renegotiate the PSN binding? Takes away freedom to map to a suitable PSN, because need to renegotiate if current PSN not OK. Ping: Today the binding is automatic and don't want to break existing implementations so don't want to take that away. This is optional so it defaults to working as it does today. Andy: do the respin of the draft and then send a request ot the list to make a WG document. 15 min - PW endpoint fast failure protection - Yimin SHEN http://tools.ietf.org/html/draft-shen-pwe3-endpoint-fast-protection-01 Yimin presented the slides Greg : from diagram you do have PW redundancy and on top of that preprovisioned local repair is this not overkill? Yimin: need backup PW from CE1 - CE2 traffic in both directions Greg: stable or transit state Yimin:may be only for a certain amount of time Greg: if you have only AC affected then LDP session not effected. Yimin: if PE2 fails Greg: why not use PW redundancy Yimin: Greg: need to coordinate local protection time and end to end or strange race conditions. Yimin: it takes time for global repair to kick in Greg: you monitor locally, but assume end to end as well. need to coordinate them Matthew: no different than using an LSP protection mechansims Jie: does this need to swap PW label, there may be another draft that solves this Yimin: No, PE4 changes the label. If PE2 fails all goes on the bypass to PE4 someone form E///: looks like IP FRR, should this be discussed in Rtg WG. Dealing with quite similar ideas Yimin: local repair could be LDP or RSVP protocol, can reuse a large part of the technology. Bypass could be signaled or provisioned Samy: Now the PLR label stack has the transport label and PW label that P2 expects. Isn't it missing something? Yimin: bypass LSP label is non reserved. At PE4 it uses a context specific label that must be signaled Lou Berger: creating a PW protection mechansim overlayed on a protection mechanism. Yimin: topology will take advantage of a protection LSP Lou: regardless of how that LSP was set up. Looks like you are binding the PW control to the tunnel control. This group has done a lot to separate them. May be better to separate them out. Yimin: Can be viewed from transport perspective. From that perspective can to local reroute and repair. Lou: worse. In order to do PW protection you need to bundle into LSP protection. Yimin: this is about PW endpoint protection Lou: you are doing PW end point protection by doing magic at the LSP level, that should be separated. ??? from Huawei: What is the difference between this PW ingress and egress protection from LSP ingress and egress protection. Yimin: In LSP protection the egress doesn't understand inner label. Senad ALU: don't understand issue of the group. Sri: useful to state how many label entries and lookups it takes to do this. Yimin: double the labels, but context specific labels. so two look ups Sri: does this work in case of link state IGP only? Yimin: Yes Himanshu: if you have VCCV BFD you want it to fail so it can switch to a backup PW, but in this case it would continue to work. Yimin: yes John Drake: If the data packets go to PE4 the OAM packets will be lost. Denis Cisco: What was already here vs what is new? May run different dual homing protocols, which puts traffic at PE3 and you end up at a disconnect. You need to handle from a CE perspective. Yimin: Both ACs and PWs are active DenisCisco: struggling with value Yimin: value is fast reroute Denis Cisco: seems like overkill George: how do you differentiate the failure of PE2 vs. and failure of the link between PLR and PE2. Yimin: failure of link treated as PE2 fail George: could have traffic come back around Yimin: before that happens control plane converges George: if traffic comes around before that you end up with reordered data. Andy: take it to the list for issue resolution before consideration as WG document. ********************************************************************** REMOTE INFORMATION FOR THE PWE3 SESSION(S) ********************************************************************** All Remote Participation Info - http://www.ietf.org/meeting/81/remote-participation.html#webex