Minutes IETF116: grow: Thu 04:00
minutes-116-grow-202303300400-00

ReHash Agenda

Chairs slides

Paolo Lucente

• draft-ietf-grow-bmp-tlv

  • Paolo Lucente presented slides

  • discussion on bumping version number

    • Job Snijders -- cost associated with bumping version number,
      authors need to consider trade offs and impact on other
      implementations.
    • Paolo Lucente -- pmacct implementing v4, FRR already
      implementing version 4, and co-author draft implementing
      version 4, others not quite there
    • Job Snijders - take a survey on mailing list in favor of
      version 4
    • no comment specific on bumping version, take discussion to
      mailing list
    • Robert Raszuk -- add-path flag specific to address-family &
      peer, not just peer

• draft-lucente-grow-bmp-tlv-ebit

  • Paolo Lucente presented slides
  • Paolo Lucente - draft is done
  • David Lamparter -- already imlement metric specific stats type.
    cannot remove from iana registry because old code. iana --
    "reserved previously used for experiment"
  • Job Snijders - version 2 says "document makes no changes to iana
    registry"
    • Job Snijders - need to update version 2 to reflect what do
      do about iana

Camilo Cardona

• draft-ietf-grow-bmp-yang-01

  • Camilo Cardona presented slides

  • Job Snijders -- upto authors on how to handle issues. some use
    github/pull request, others use mailing list requests.

    • should update group about status of changes

  • no questions

• draft-cppy-grow-bmp-path-marking-tlv-12

  • Camilo Cardona presented slides
  • Job Snijders -- send email to start process to adopt
  • Job Snijders -- why is reason code only 2 octets. many
    datapoints to check. ebgp transit provider 10-20 exit points
    result in reject for route policy. 64 points seems limited.
  • Camilo Cardona -- 65k reason codes, status codes only 64.
  • Job Snijders -- lets make sure we have ample room to give
    operators flexibility.
  • Ben Maddison -- usecase aproximate instrumentation of route
    policy, expecially in test. make reason/status code wide enough.
    add hierarchy. may not have vendors that can make it work.
  • Camilo Cardona -- route-policy draft (need specific reference)
    add complexity of header key, 2 are complimentary. this is
    cheaper way of covering it, would prefer to keep this one
    simple.
  • Ben Maddison -- 2 drafts are same thing. not worth having both
    drafts.
  • Camilo Cardona -- proposes one option remove reasons code, and
    keep status here, and then use other draft complex way.
  • Ben Maddison -- camilo's idea could work. just status code
    without reason not useful. camilo usecase just need status.
    battling with complexity of adding scope. ben - have
    intergration tests that try to cover routing policy. routing
    policy is super complex, and wants to perform coverage test of
    route policy.
  • Paolo Lucente - path tracing draft no longer active, very
    complex, implemented it in collector. now have REL draft
    presented at 115, and cover many of the usecases. looking for
    input on REL. Please provide input!
  • david -- distinction between trace and heirarchical
    status/reason, trace too expensive, but status/reason useful.
  • Ben Maddison looking for error chain rather than trace
  • Job Snijders -- want to correlate exit point of routing policy
  • Job Snijders -- mail sent to list ask to adopt
  • Rüdiger Volk -- coding trace through route-policy, implementors
    may not be able to code. wants drop reason code on RIB-in,
    suggestion "allow in config language drop statement support
    integer", not no complex coding
  • Jeff Haas -- when you want to output data 2 choice: spit out
    immediately or hold in state. stack trace is difficult,
    proceedureal state machine, gdb style stack trace, have to
    maintain stack, item 3 how to represent in fashion that is
    useful. do you want internet route's worth of stack trace eating
    memory. things are doing able but expensive. just note thing is
    rejected, have a catalog of reasons, netconf operation, test
    route, instead of state
  • Ben Maddison -- tests run not on production network. test
    environment like production. likes Rüdiger Volk idea, and extend
    this to many action statements. jeff haas large integer
    required, have to hold onto code until installed.
  • Jeff Haas -- I suspect if you thought about how many bits it
    would take to actually represent the state that you're looking
    for, good find that your scaler is actually a very very big bit
    string.
  • Rüdiger Volk - at some point vendors implemented instrumentation
    for policy implementation turn on statistics "like give policy
    visited and how much time" maybe useful if still around
  • Job Snijders -- continue on mailing list

Pierre Francois

• draft-francois-grow-bmp-loc-peer-00
  • Pierre Francois presented draft
  • Jeff Haas -- vbit redefined. make this feature difficult as
    existing implementations already expect. Loc-RIB view sent
    initially (peer up state) optional TLV RIB name. expectation of
    name of table. peer up message include system-id. add additional
    TLV for peer up, station up keep track. can keep it in each
    route montioring message, or just have it at peer up message.
    That's more for the case where IPv4 you were correct. For the
    case where it is IPv6 , the v-bit is otherwise move for
    something else, you would be changing the parser path for this.
    I believe.
  • thomas -- if not part of RMM just shifting data downwards
  • Pierre Francois -- cannot keep it per a peer with add-path, so
    becomes quite complex
  • Camilo Cardona -- make sure yang model supports it
  • Job Snijders -- RFC9069 -- not strong language to confirm the
    field is all zeros. per peer header ipv6 16B of zeros, and ipv4
    4B of zeros
  • Job Snijders -- how did this situation come to be? oversight?
    some implementation struggled to get information.
  • Pierre Francois -- own implementation more difficult to make
    zero than set to peer address
  • Paolo Lucente -- rfc9069 author, what francois said above
    vendors struggling so just make it zero field
  • Jeff Haas -- 7854 first is vbit second is fbit. are you making a
    new field?
  • Pierre Francois -- just use field right after
  • Job Snijders -- doesn't seem new bits need to be added
  • Jeff Haas -- rfc7854 per peer header address can be very long in
    size dependant on vbit. rfc9069 same bit position different name
    space. must be changes in impelemtation loc-rib peer, or rib-in
    peer
  • Job Snijders -- some implementors may ignore v-bit. not vbit for
    loc-rib peer, reusing field loc-rib/rib-in
  • Job Snijders -- take an unused bit to prevent issue
  • Jeff Haas -- rfc9069 behavoir should be new version number
  • Job Snijders -- may need TLV
  • Jeff Haas -- TLV is fine, but not longer "free change". If
    collector already knows who it talking to, knows RIB, optional
    TLV for system address/router-id. collector has enough
    information to decorate db
  • Paolo Lucente -- preference to BMP stateless BMP TLV. make BMP
    as stateless as possible. reduced complexity, stateless. first
    fbit stay, 2nd unused (0). does see problem with using existing
    bits
  • Job Snijders -- issues not insurmountable.
  • Pierre Francois -- needs update to rfc9069 (job says yes)
  • Job Snijders -- show of hands, interested in continuing this
    work -- 10 or so support work
  • Job Snijders -- take feedback, make decision which direction
    (TLV or bit), upload update, then consider working group item