Minutes IETF114: icnrg: Thu 10:00
minutes-114-icnrg-202207281000-01

Meeting Minutes Information-Centric Networking (icnrg) RG
Date and time 2022-07-28 14:00
Title Minutes IETF114: icnrg: Thu 10:00
State Active
Other versions markdown
Last updated 2022-07-31

minutes-114-icnrg-202207281000-01

ICNRG & COINRG

Joint Meeting @ IETF 114

ICNRG Chairs

https://datatracker.ietf.org/doc/slides-114-icnrg-chairs-presentation/

  • List of active documents presented
  • CCN info to be completed
  • Path steering for RG document? Follow up on the list - but the IPRR
    for IRTF is non-existent - not a standard - 3rd party disclosure by
    Dave.

From the chat window:

  • Lixia Zhang: I believe the very initial idea (of using an Interest
    to set up path for another interest) came from Yu Zhang when he
    visited UCLA. Wonder whether the IPR application ACK'ed that fact.
  • Dave Oran: Well, for 3rd party declarations you only say what you
    know about, not what others may know about. It would be up to Yu
    Zhang to put one in. Is there anything written down? I'd be
    interested.
  • Lixia Zhang: Yu Zhang has an ICN poster some years back, also an NDN
    Tech report at the time.
  • Dave Oran: Thanks I'll try to dig this up - the stuff in the draft
    comes from Ilya & my ICN'17 path switching paper, so the patent
    application predates that by a few months or so.
  • Lixia Zhang: I just dug out the poster info:
    Poster: "Kite: A Mobility Support Scheme for NDN"
    Yu Zhang, Hongli Zhang, Lixia Zhang
    ACM Information Centric Networking Conference, 2014.
  • Dave Oran: Oh, I know about Kite. It doesn't force packets onto a
    path, it just ensures the forwarder state is there.

Ping Traceroute - Dave Oran - (update)

Alternative Time Encoding - Thomas Schmidt (update)

From the chat window:

  • Carsten Bormann: It is pretty easy to pick apart a float32 and get
    this.

Selective Content Disclosure Nikos Fotiou (research)

Closing Comments from the ICNRG Chairs**

  • FLIC needs to move forward

  • New work requested:

    • Media over ICN (like is done for QUIC)
    • Self learning/autoconfiguration/switch design
    • Web over ICN

  • Joint work with COIN?

  • Lixia Zhang: (1) What are the key problems needing solutions? Why do
    you do that? (e.g. what problems are identified from running media
    over QUIC? That may be overcome by running media over ICN). (2) the
    basic idea of ICN is networking via named, secured data. Wish
    more ICNRG efforts look into the use of sematic naming, and
    solutuions to address security challenges.

    • Dirk Kutscher: Some of these points are related to real problems
      e.g. real time distribution over the Internet. Security and web
      over ICN are pressing issue.

COINRG Part

COINRG Chairs

Traffic Steering at Layer 3 - Dirk Trossen (research)

  • https://datatracker.ietf.org/doc/slides-114-icnrg-traffic-steering-at-layer-3/

  • IFIP networking paper:
    https://doi.org/10.23919/IFIPNetworking55013.2022.9829778

  • DirkK: Why call this "semantic routing"?

    • DirkT: it's actual "service routing" – "semantic routing" is
      just a relict from previous work

  • KenCalvert: What is the advantage of doing this @ L3 vs L7? I think
    this paper has looked at doing something similar @ L7: E. W. Zegura,
    M. H. Ammar, Zongming Fei and S. Bhattacharjee, "Application-layer
    anycasting: a server selection architecture and use in a replicated
    Web service," in IEEE/ACM Transactions on Networking, vol. 8, no. 4,
    pp. 455-466, Aug. 2000, doi: 10.1109/90.865074.

    • not average latency but variance of latency (jitter)
    • variance would be reduced
    • other scenario: resilience, e.g., when overloading servers
    • in L3, you can distribute over all clients and thus get better
      performance
    • The IFIP Networking paper:
      https://doi.org/10.23919/IFIPNetworking55013.2022.9829778

  • DirkT:

    • @Kenneth sorry, only now realised your question at the bottom of
      the reference. Yes, the paper outlines the CDF to show the
      lowering of the variance through CArDS compared against the
      other mechanisms.
    • @Kenneth, also note that CArDS only requires signaling the units
      used at deployment time, no regular metric reporting is
      necessary (such as suggested in the reference you sent)

Namespaces - Andy Reid (research)

Building Adaptive Networks - Tushar Swamy (research)

  • https://datatracker.ietf.org/doc/slides-114-icnrg-building-adaptive-networks-with-machine-learning/

  • Need to do ML fast to keep up with the dataplane; lead to Taurus
    switch pipeline design to supply line-rate ML inference. A
    programmable fabric. Published in ASPLOS.

  • How do we program Taurus? Led to Homunculus project, a framework for
    dataplane model generation. Homunculus provides net operators a
    high-level compiler to program ML switches.

  • See the last slide for a URL pointer to an archived version of the
    paper that is under submission

  • Take-away: an adaptive feedback look for network adaptation

  • Taurus: https://dl.acm.org/doi/10.1145/3503222.3507726

  • Homunculus: https://arxiv.org/pdf/2206.05592

  • Try it out! https://gitlab.com/dataplane-ai/taurus

  • Eve Schooler: the ML needs to be at line rate. That seems to be one
    of the challenges here – more than match-action. What kinds of
    feature extraction are targeting? What about image data?

    • Tushar: haven't done this yet. Looking to start dealing with
      image classification in the data plane, but not certain what
      would be the best use case for motivating this. System is still
      fairly resource-limited on the switch.

  • Hesham ElBakoury: does the model get updated in the control plane
    (not only in the data plane)?

    • Yes