------------------------------------------------------------------------------- MPLS WG Thursday, November 17, 2016 (KST) - 15:20-17:50 Thursday Afternoon session II Minute taker: Rakesh Gandhi ------------------------------------------------------------------------------- 1. Agenda bashing, WG status, reports Presenter: Chairs Time: 15:20:00 - 15:40:00 - 20 minutes Slides: We had a Yang meeting with CCAMP on Monday afternoon. Today MPLS specific drafts. Note well presented. No errata have been reported. Many thanks to Eric ? for Liason document work. We have two new RFCs, RFC 7965 and RFC 8012. We have 4 new WG documents + 1 more for mLDP yang draft. draft-3107bis should be in a good shape. Out of 20 WG, about half of them have been updated. mpls-tp-aps -> in the WG LC. mpls-opp-encrypt -> currently being implemneted at some univ. Make exp RFC? mpls-mldp-mib -> Read only document. Beniout is OK to porgress this one. mpls-ldp-mrt -> stuck in the Sheferd's writeup. Check other WG document for MPLS impact. shared-ring-protection document: Eric -> Status back on H plate. Discussed the details on the changes. Then Sheferd writeup. Their review as part of WG LC. George: Basically just need to update the doucment. All changes agreed upon. RFC4379bis: Approved by the ISG. IANA started to take actions. Coming along fine, don't see any problem. draft-xxxx Deborah can send it back - AD watching document. Base Yang model: ready for WG LC prior to chichago. Residence time: submitted to ISG for ppublication. New IDs: Label format change, moving bit around; we might do it and need some presentations on it. mpls-ldp checnpen: ready to present in Chicago. Kireeti need to rename draft to mpls-ldp and mpls-rsvp. Progress Base+ Static: Some small things to do, warning and errors to fix. static to add p2mp support mpls-summary-frr: outstanding review, Shepherds review. promised to come back with a response soon. --------------------------------------------------------------------------------- 2. draft-bryant-mpls-rfc6374-sfl Presenter: Stewart Bryant Time: 15:40:00 - 15:50:00 - 10 minutes Slides: This update is for delay and jitter. Runs on the synonymous label, OAM message disgused as real msg. Loa: Can you explain synonymous label? Stewart: Label with twin characteristics. Special label between src/rec. Far-end treats exactly as original label. Count number of packets for example. Kireeti: Does it not bother you this packet may be load-balanced differently. Stewart: if this is an issue, use entropy label system. George: There is a lot of h/w that counts packet by label. Don't have to buid new h/w. Stewart: Little bit of control plane using existing h/w. Use third syn. label for delay measurement. Request feedback on what combinations of loss /delay need to do. Always measure loss and then sometimes delay meas. Method 1: setup some bins/buckets for delay meas. Method 2: classic standard deviation over a batch. Method 3: Average delay - one way delay measurement. Sum of timestamps. Next steps: Need some comments on measurement preference. Get this right and then go back to the archi. Draft is not perfect but need to solve the problem. Tal M: Read draft. valuable to the WG. Loss and delay use 3 syno, labels. Ste: yes. One loss Meas. Sub meas - use a pair for delay. prob. need 4? George S: Delay meas is an OAM, 2 other labels alternate between two path of couting. Stewart: We are not quadrupling the number of labels needed. Tal M: Multple marking draft was presented. Same conept can be applied here. use two syno labels. If you make assumption about certain guard bands. George S: Syno. label A, B treated exact same thing. We don't want to mix OAM in there. Stewart: protocol work. How you build the appl is upto you. Tal M.: You are not dis-agreeing withm me? Stewart: Not convinced it is needed to minimize the number. Loa: Send an email to the list. Authors contact him and discuss, Stewarts: Will add some text for both methods. Loa: IP PPM WG socialize. George S: ? Stewart: Can we make this a WG doc? Loa: Don't see a reason not to do it. I need to read it. George S: Many way to do DM but this is a better option. Stewart: We will work with meas expert. Design payload. Minimize the number of options. Loa: who read the document (fair number) Loa: who thinks we should WG adoption (about the same) ----------------------------------------------------------------------------------- 3. draft-xu-mpls-service-chaining-00 Presenter: Stewart Bryant Time: 15:50:00 - 16:00:00 - 10 minutes Slides: SFF: service function forwarder. Next steps: we need to look at eignal Which WG ? Loa: need to discuss this. Stewart: This may an applicability statement. Himamnshu: there is no new protocol. Stephane Orange: One concern. Just by doing SRTE, Trouble with label stack today. Adding more label stack? Stewart: There are other techniques we can use. Stephane Orange: Add states in the network. Stewart: Indeed. Loa: Problem with h/w - number labels. Only problem at the ingress? Stewart: Yes. S/w puts label stack for you. Jeff T: For push part number of the labels - PCE draft for SFC and SR, computed path. Binding sid concept should be looked at. Stewart: Indeed. Piggy back on SR. Jeff T: BGP-LS extension. SPRING right place? Stewart: ?? Himanshu: Number of labels - not a concern. Jeff T: Stuck with label stack limit. Cavor Rose ??? : Concern about large number of labels. Intensive CPU wise. George: Need to look at the bottom stack. Cavor Rose ??? : Hurt performance. Stewart: Least of your problem. Cavor Rose ?: Ability to inject traffic. Stewart: VLAN, Ethernet, cable. Lucy Yang Huawei: Yoy are assuming that classifer will select one of them. Stewart: Just like anycast SR label. Labels to identify function and instruction. Jeff: Cumbersome operations. Rebuild the packet and it is problematic for MPLS encap. Adrian: Thank you for this draft. Some good ideas. Similarity with some other work. BESS Agenda adverisizing SFF in BGP. May be look at that. Stewart: Can work with you on this. ------------------------------------------------------------------------------------ 4. draft-xu-mpls-spring-islands-connection-over-ip-00 Presenter: Xiaohu Time: 16:00:00 - 16:10:00 - 10 minutes Slides: Discussions on do we need this draft ? George: Personal opinion - UDP draft is sufficient. See 3031. have mechanism to send MPLS over IP. Stewart: This is an area we really need to get hands on. ------------------------------------------------------------------------------------ 5. draft-esale-ldp-rmr-extensions-00 Presenter: Kireeti Kompella Time: 16:10:00 - 16:20:00 - 10 minutes Slides: Loa: Question LDP is multi-point - disable this fucntion in ring? Kireeti: No. LDP is MP and no issues with LDP. RSVP is P2P which may be an issue. Kireeti: Look and think about, improbvement and please send to the list. No questions. -------------------------------------------------------------------------------------- 6. draft-ietf-mpls-rmr-02 Presenter: Kireeti Kompella Time: 16:20:00 - 16:30:00 - 10 minutes slides: Architecture document. You have a notion of a ring neighbor. Clockwise neighbor and anti-clockwise neighbor. George: Express link also clock-wise? Kireeti: Details not worked out. Used to load-balancing. They are neither and they are both i.e. clock-wise and anit closewise. Path computation to best way to traffic for express links. Stewart: Express link is another element of the mesh? Kireeti: Yes. Like a trench. create express path when lot of traffic. Stewart: Express link as part of the protection mechanism? Kireeti: How to do basic ring for protection. Then add express links. Express neighbors. George: Kireeti: Express link for opimization. If link fails, what do I do with traffic Don;t want to create a loop. Seconds thing is when that happnes, convrge to a smaller ring. Express link allows that. Stewart: ?? Kireeti: In this situation, you often have multiple failures. Stewart: Ring wrapping to fix ? Kireeti: Yes in some cases. Careful not to cause loop in the process. Stewart: Let the IGP sort it out. Kireeti: Yes, slowly. Stewart: Use SR. Kireeti: Yes. IGP sort it out - not true. IGP extensions defined. Details in IGP drafts. No questions. -------------------------------------------------------------------------------------- 7. draft-deshmukh-rsvp-rmr-extension-00 Presenter: Kireeti Kompella Time: 16:30:00 - 16:40:00 - 10 minutes Slides: RSVP signaling after IGP discovery. Ring is MP2P. George: All mech are already in RSVP. Kireeti: If you have comments on our proposal please let us know Kireeti: New C-Type for RMR Tunnel in Session. Need for V4 and V6. Have hop-by-hop new BW - TSEPC. We considered having multiple sender templates. ? : What kind of algorithm for bw, first come first serve? Kireeti: Yes. Except existing reservations are honored. Number of items under study e.g. toology of the ring changes, traffic impact while ring converges. How to use express links? How to protect traffic over express links? ------------------------------------------------------------------------------------ 8. draft-esale-mpls-ldp-node-frr-04 Presenter: Kireeti Kompella Time: 16:40:00 - 16:50:00 - 10 minutes Slides: Kireeti: Draft is using RSVP. Can also use SR. Stewart: Potential confusion for RTGWG? Chris Bowers: RTGWG happy to be over there. Works for LDP, now including RSVP, might be good audience. Stewart: Socialize in both places RTG and MPLS. Stewart: Is this documenting what is already well known? Kireeti: Yes. Loa: Is this documented somewhere else? Geoerge: Not documneted at IETF. Do we need to? Kireeti: Is this BCP? --------------------------------------------------------------------------------- Note takers: Rakesh Gandhi, ---------------------------------------------------------------------------------