PWE3 - Wednesday, March 25, 2009 13:00-15:00 --------------------------------------------------- CHAIRS: Stewart Bryant and Matthew Bocci 20 min - WG Status - Stewart Bryant and Matthew Bocci Stewart opened the meeting. Matthew presented the status. 6 documents in Editor's queue. Congestion framework on hold. Please read the documents that are in last call and provide comments. Chairs will be following up with last call documents after the meeting. For Multsegment PW Architecture, because both chairs are authors, the Secretary, Dave Sinicrope, will call consensus on the draft. Note that vccv-bfd is not on the agenda although in the WG Status slides. 10 Min - Pseudo Wire (PW) OAM Message Mapping - Peter Busschbach http://tools.ietf.org/html/draft-ietf-pwe3-oam-msg-map-09 See posted slides. Peter presented. April 2003 is when Monique and Tom published first message mapping draft... we are finallly where we need to be at this point. Major Changes Slide 2 Last gap has been filled for TDM and CEP PWs. The CEP section needs review and needs input from the experts. Minor changes slide 3 Confusing terminology for reading, changed to use receive and transmit terminology. Next Steps Slide 4. Authors believe document ready for WG last call. Q&A - none The chairs will make a last call on the list. 10 Min - Segmented Pseudo Wire - Luca Martini http://tools.ietf.org/html/draft-ietf-pwe3-segmented-pw-11 See slides, Luca presented. Slide 2 Could change the fault terminology to defect instead of fault. Make comment to list if desired. Slide 3 Stewart had requested to refer to MS-PW arch document. Luca kept terminology in the document for readability. No problem duplicating, but terminologies should be word for word identical. Don't need to remove the intro section just shorten it. Proposal to remove L2TPv3 section. Question was who is interested in switching L2TPv3 PWs to MPLS or other L2TP PWs? Little interest in L2TPv3 PWs so less interested in interworking. Greg: Would have thought that L2TP MS PW to MPLS MS PW interworking would have been most interesting part. Luca: no interest Greg: don't block in future Luca: OK can separate it out or leave it in here. Stewart: Only raised as a review comment because majority not interested Mark: Isn't this text already in there? Carlos might be able to adequately review so what is point to removing? Luca:Carlos reviewed Tom: There was concern that if left in the IESG would have a problem with it. Agreed to leave the section in. Slide 4 - Next Steps Finish LC comments - need to discuss some with Stewart. Want to process the MS-PW arch and segmented PW documents together. Bhupesh: There are 3 control channel types. Type 2 can't be used for end - end, Type 3 is the TTL type and don't know what value to set. Type 1 is left for OAM, and has optional control word. How is the OAM for MS PW expected to work when the different types all have issues? Luca:Control word is required if you want function, so allows for no VCCV support. TTL is spelled out in document. Could use a Type 1 and still set the TTL or type 3 and set TTL. Not removed because some implementations do this. 10 min - Application of Ethernet Pseudowires to MPLS Transport Networks - Tom Nadeau http://tools.ietf.org/html/draft-ietf-pwe3-mpls-transport-03 See slides, Tom presented status update. Document expired and has been delayed due to effort spent on MPLS-TP Document not meant to replace MPLS-TP work, this is a half step. This document goes a lot towards meeting requirements although doesn't go as far as MPLS-TP. e.g. OAM Document in last call, will end in 2 weeks. Folks interested in MPLS-TP should read and comment. Note: Document will be informational. 10 Min - Packet Pseudowire Encapsulation over an MPLS PSN - Andy Malis http://tools.ietf.org/html/draft-bryant-pwe3-packet-pw-00 See slides. Andy presented. Goals Keep in mind: Intended to be used with MPLS-TP. Observation is that number one application is router interconnection over a transport network. Router interconnection requires transport of several protocols. Aiming for the case where the client LSR and server PE is the same piece of equipment. Reference Model Slide NOT trying to support case where Client and PE1 are separate equipment. Not just MPLS slide Interconnecting routers presents more than just IP between routers. Can be done with e.g., PPP or Ethernet. Simpler Solution could do with Ethernet, but fairly complex for just interconnection. Packet PW Ref Model Using a PW over MPLS TP to interconnect light colored boxes. Note that the PE and LSRs are in the same physical box. This accomodates several shortcuts. Packet PW Control Word All fields from PW control word along with PPP PID. Status Indication. Because virtual interface need to map the status across to the remote side. Questions Florin Balus: Slide Packet PW Control Word - Don't you need the first four bits to be different from PW? Andy: no, it is just a PW Riccardo Martinotti: Is it possible to decoupled to be able to use encapsulation method in other situations. Andy: you would then need to add quite a bit of complexity and we already have PWs defined that can handled the complexity. It's a solved problem. This is an optimization to handle this case. Stewart: adjacency formation is more complex than folks think. This is a special case, we could simplify but for other cases use PW. Roman Solomon: What is advantage to using this vs a PW to a virtual interface. Andy: packet efficiency Stewart: also control plane is a bit different Roman: Back to Packet PW Ref Model - This could be an ATM cross connect. Implementations currently there using a virtual interface with an ATM PW. Could maintain that model in this draft. Andy:In this case you have you would need to run full ATM stack across transport network at 40 or 100 G. Roman: it is running Andy: take to list Kireeti : is this only for adjacency formation? Andy: yes Kireeti: why do you have a 2 byte PPP id? why not go to 1? Andy:don't want to start registry Kireeti: How do you know if you should forward or pass traffic to the control plane? Andy: same way done today on a router Kireeti:then not just for adjacency formation, really an accidental by-product Andy: router interconnection over MPLS-TP. Kireeti: full encpasulation type like PPP only simpler Greg Mirsky: mentioned UNI in this model is between PE and LSR passing MPLS? Andy: on this UNI you have MPLS, IP router or anything else Greg Mirsky: clarify what it will carry Andy: carries anything that PPP can carry. Would like to test consensus to make a WG draft. Read draft: 30 Good way forward: ~25 10 Min - Flow Aware Transport of MPLS Pseudowires - Stewart Bryant http://tools.ietf.org/html/draft-bryant-filsfils-fat-pw-03 See slides, Stewart presented Merged text that Vach and Shane did. Stack slide: Flow label must be greater than 15 so not interpreted as a reserved label. Signaling Up to PEs to craft the label above the reserve labels Ability for one end to run flow label and other not, is a silly idea so want to cut it out. Kireeti: Why is it silly? why remove it? Stewart: much more complex signaling Kireeti: much harder to do ingress imposition of a flow label than pop it at the end. It is possible to get benefit in one direction. Stewart: the signaling was more complex than needed and broke symmetry Luca: Don't have to put anything useful there, can put same label each time if you don't want to use it in other direction. Yaakov Stein: must be bidirection because a PW is bidirectional Concensus - put useless label in one direction OAM issue Yaakov - problem not having OAM. If you're running you'll see packet loss vs. failure Stewart - would use FRR and not really have to be concerned Yaakov - I use VCCV because I do want to see failure Stewart - primary use of this is for internet flows Kireeti - agreeing with Stewart Single Large Flows Yaakov: Put the option in the document Luca: Not going to prevent using random label Stewart: just negotiate a different option Make sure that we don't affect hardware, so please look and comment Should remove the option to run without a control word Yaakov: should we go toward a model where we don't allow to run without a control word Luca: 99% is ethernet which doesn't use the control word Yaakov:99% of internet is v4, doesn't mean should stay with it forever Stewart: test concensus Read draft: 13 Ready for WG draft: 9 Take to the list to see who else has read it. 15 Min - LDP Extensions for Source-initiated Point-to-Multipoint Pseudowire - Yuji Kamite for Fred Jounay http://tools.ietf.org/html/draft-jounay-niger-pwe3-source-initiated-p2mp-pw-02 See slides, Yuji presented Encourage WG to look at document and provide feedback. Greg Mirsky - VPMS allows only one attachement circuit at ingress Yuji - would have more than one connection to keep reliablility Greg Mirsky - could be more than one pw, but only one active Read draft: 8 Need to take further comments on the list. Yuji - is it reasonable to drive this via source initiated first? Need to take to the list. 10 Min - Inter-Chassis Communication Protocol for L2VPN PE Redundancy - Luca Martini http://tools.ietf.org/html/draft-martini-pwe3-iccp-01 See slides, Luca presented. Several people already in WG working on the document, so much interested. Read:20 Ready for WG draft: 16 Looks like consensus to start WG draft, but Stewart will check with the list. 10 Min - Encapsulation Methods for Transport of InfiniBand over MPLS Networks - Vikas Puri http://tools.ietf.org/html/draft-puri-pwe3-ib-encap-01 See slides, Vikas presented. Stewart: questions - can infiniband stand the packet loss you get in an MPLS network Vikas:Do you mean attachement circuit failure? Stewart: No, PSN error Luca: is the lossless on the introduction slide mean requirement or what infiniband provides Vikas: the PW sould be transported over a a TE LSP if possible, but there are mechansims that are part of the protocol to handle loss. These are not initially proposed. Stewart: Worried about infiniband congestion the PSN network. How would you avoid overloading? Vikas: Infiniband has specific bandwidth features and rate limits. Need to explore what an operational network looks like. CAC and signaled bandwidth limits will be explored. Stewart: so you envision this to be run on a fixed bandwidth RSVP-TE bandwidth reserved network Vikas: yes ???: pseudo port which becomes the pw, the rate of the pw becomes the port rate that the subnet manager on the IB sees. So if we know what that rate is then we can signal it. David Black: reinforce Stewarts point. Take a look at the fiber channel PW. Opportunity to view as more than a fixed rate connection. Luca: There is a FECN concept so this is what we are looking for as a way to tell a host to slow down. Vikas: yes David:opportunity to do dynamic reaction to what is going wrong. Stewart: is this your proposal or the infiniband org proposal Vikas: this is our proposal Read: nearly 0 Need to ensure that the Inifiband group are happy with us working with/on their protocol. Luca: please send email to list introducing document to the list. Stewart: although minority interest doesn't mean we shouldn't help them. 0 Min - Ethernet PW explicit congestion notification - Yaakov Stein (no draft) See slides, Yaakov presented Limiting to Ethernet PWs used for cell backhaul for 3G with large data volumes. Would like people to think about mechanisms. We have control word that is sitting there. Just need 2 bits for the flexibility. Howard Green: already a draft to extend PCN marking to MPLS. Could affect this. Stewart: That is a bit different because it tells the MPLS clients what to do, but this is in client layer itself. Howard: There is a difference but also a relationship Stewart: This is a bit different because no impact on MPLS network. David Black: avoid ECN acronym collision Yaakov: can call it a BECN Stewart: is this an area that we should be encouraging Yaakov to work on Should we be working on this? No opinion David: this is one mechanism, but need to work on big [congestion] picture Stewart adjourned the meeting at 3:02pm PDT