IETF 121 MPLS WG Meeting

Date/Time: Wednesday Session II, November 6, 2024 13:00 - 14:30 (local)

Room: Liffey Hall 1

Chairs: Tarek Saad/Nicolai Leymann/Tony Li
Secretary: Mach Chen

Slides:
https://datatracker.ietf.org/meeting/121/session/mpls/
Codimd for Notes Taking:
https://codimd.ietf.org/notes-ietf-121-mpls/
Meetecho:
http://www.meetecho.com/ietf121/mpls/
Zulip:
https://zulip.ietf.org/#narrow/stream/121-mpls

  1. WG Status Update (Agenda Bashing) - 13:00
    Duration: 10 mins
    Presenter: WG Chairs
    [Greg]: we reached out to the other authors of the IOAM-dex
    documents and we are working on collaborating/combining into one
    draft.

  2. Stateless MNA-based Egress Protection (SMEP) - 13:10
    ID:
    https://datatracker.ietf.org/doc/draft-ihle-mpls-mna-stateless-egress-protection-00

    Duration: 15 mins
    Presenter: Fabian Ihle
    [GregM]: did you think about local protection too? This may be
    applicable to SR-MPLS (for example)
    [Fabian]: It is possible, but for transit protection many other
    approaches
    [Sasha]: encoding mentions 1 label but do we need to specify the
    interface on PLR? This could work with SR.
    [Stewart]: on genralizing the idea, need repair for every hop and
    that does not scale. An alternative, is to have a different action
    (POP-N). Does not necessitate rebuilding the MPLS stack.
    [JieD]: encoding, ... Scope, how do you control which PLR to
    process. If there are multiple PLR, then multiple NAS with repair
    stack for each.
    [Fabian]: yes.

  3. Update on MNA Implementation Experience in P4 - 13:25
    Duration: 20 mins
    Presenter: Fabian Ihle
    [Jag]: is there any difference between the implementation between
    version-5 and version-8
    [Fabian]: we implemented version-5 and it worked. Latest version
    can carry AD for format B LSE
    [Rakesh]: we can discuss if it makes sense to shorten the NASL
    length
    [Greg]: one of the concerns I have when putting AD in PSD makes
    challenging for parsing.
    [Fabian]: correct. For PSD, we need 51 LSE NAS + PSD. Even if
    using 2 LSE (and P bit) to indicate PSD may be lots of overhead.
    [Greg]: if we have multiple MPLS payloads (e.g. PW or BIER), then
    we can push the MNA PSD even deeper.
    [Rakesh]: In the MNA-HDR solution, we have a flag and offset. Does
    it help to reach the data?
    [Fabian]: from P4 implementation, it makes this complicated. Not
    so useful.

  4. Performance Measurement Using Simple Two-Way Active Measurement
    Protocol (STAMP) for Segment Routing Networks - 13:45
    ID: https://datatracker.ietf.org/doc/draft-ietf-spring-stamp-srpm-16

    Duration: 10 mins
    Presenter: Rakesh Gandhi
    [GregM]: the loopback mechanism is useful when no clock
    synchronization between 2 nodes. You can divide by 2 to infer.
    [Tony]: since this depends on the MNA-HDR draft (which WG agreed
    will move to experimental), this document may need to move to
    experimental.
    [Rakesh]: we need to investigate if an informative status would be
    impacted by referencing an experimental doc.

  5. Label Switched Path (LSP) Ping for Segment Routing (SR) Path Segment
    Identifier with MPLS Data Planes - 13:55
    ID:
    https://datatracker.ietf.org/doc/draft-ietf-mpls-spring-lsp-ping-path-sid-02

    Duration: 5 mins
    Presenter: Xiao Min
    [Greg]: Does the PCE document ensure that the allocated SIDs at
    egress are known at the ingress?
    [Xiao]: if path-segment is not used in the network
    [Greg]: what happens if the headend is not aware of the state
    present at the egress? Both specs are normative.
    [Ketan]: Path-segment is a label that belongs to the egress (from
    its label space). Some mechanism is needed to make this known to the
    headend. I think we should progress this work independent how state
    is reported from egress to ingress.
    [Greg]: the candidate path is selected by the ingress. If we can
    not synchronize state between ingress and egress, then this is not
    useful. If BGP can not synchronize this state e/i, then we should
    not have it in scope. In any way, the document needs some rework.
    [Ketan]: solution does not require control plane to synchronize.
    This can be done by config/provisioning. There is another option
    using PCEP. BGP may have challenge doing this.

  6. Deterministic Networking specific MNA - 14:00
    ID: https://datatracker.ietf.org/doc/draft-varmir-mpls-detnet-mna-00

    Duration: 15 mins
    Presenter: Greg Mirsky
    [TonyL]: Lou Berger suggests that ECMP in DETNET is 'not
    recommended' not 'prohibited'
    [XueyanS]: there is some overlap for this with another document.
    [Greg]: the options can be combined for different deployments. One
    of the advantages is we can have single seq-no. space. We will be
    going to the DETNET WG to get their feedback. We are open to other
    feedback.

Total presentation time: 75 mins
Shuffling time: 15 mins