MOPS IETF 110 Virtual WG Meeting

March 12, 2020.

Admin [5 min]

• Note well
• Agenda bash
• Jabber scribe
• Minutes taker (https://codimd.ietf.org/notes-ietf-110-mops#)
  • Jake Holland volunteered

Industry News/Experiences [60min]

• Discussion -- Learning in situ: a randomized experiment in video streaming (Francis Yan) [10min]
  PREWATCH: https://irtf.org/anrp/IETF110-ANRP-Yan.m4v
  PAPER PRESENTATION: https://www.usenix.org/conference/nsdi20/presentation/yan
  • Spencer: it takes 2 years of data to measure 20% difference in performance between ABRs, is that right? (Yan: yes). OK, given that can more constrained paths be more useful, so there's congestion?
    (Spencer) And is Puffer just bound by content restriction? Yan: yes, Fugu can work on any video providers, it would be nice to extend this internationally with more than 1 server, as this would give us more scope. However, the conclusion of noisy data we believe still generalizes. The rebuffering events are so rare that it needs to be measured on tail users. Confirmed in principle by discussion with youtube and netflix.
    (Spencer): and this platform is open to other researchers, right? Yan: yes
  • Renan Krishna: In figure 6 what's the information included in the state update? Yan: in this figure the MPC controller is responsible for calling the predictor for the time of the next proposed chunk. Based on this the player can plan ahead. This diagram represents all the information needed by both components to work.
    Renan: so MPC is outputting a probability distribution? Yan: essentially, yes. There's simpler ways to look at it too, but yes. There's some scaling considerations too, so the signaling
  • Will Law: What's the bias in the test data? Is it coming mostly from the people watching this selectively have better internet connectivity? Based on other providers we usually see as much as 1% of sessions impacted by buffering, so that's at odds with 2 years of data to see differences
    Yan: Yes, I think this is consitent. We see something like 0.1% buffering on a per-request basis so it seems roughly in line. To clarify: The training takes only 2 weeks, there is enough information there to be relevant, but getting the confidence intervals low enough to be sure of relative ABR algorithm performance takes more like 2 years of data.
• SVA updates (Glenn Deen/Sanjay Mishra) [20min]
• Low Latency Streaming (Will Law) [30min]
  • Jake: best practices?: Will: dash industry forums, (tbd: come back and summarize from video, note-taker briefly lost connectivity)

WG Docs [20 min]

• Ops Cons (Spencer Dawkins)
  • Spencer RE Issue 31: Matt Stock question RE: suggestion about incorporating ads? Mike English: Yes, spoke to him, he's fine with doing what you think is best with his prior contribution.
  • Glen RE Issue 24: how does this interact with the insights about streaming? Spencer: big impact was a lot more off-peak traffic, more videoconferencing traffic during the day from home. Good workshop report, trying to minimize information here since workshop report is now a good reference.
    Glen: interesting insights on non-videoconferencing traffic in the same time space, (e.g. netflix/games). Spencer: yes, there were interesting points about this in the IAB report.
  • Jake RE volunteer request slide 6: Self-assignd a couple of issues since your slides, that's on github now.
  • Mike English to submit a PR for ads
  • Some further details of issue assignment captured in github.

Related IETF work [20 min]

• Media Operations Use Case for an Augmented Reality Application on Edge Computing Infrastructure
• Renan Krishna draft-krishna-mops-ar-use-case-02. (Renan Krishna)
  • Cullen: Latency budget is important to mention here. Renan: Sure, makes sense.
  • Leslie: what do we think of WG adoption of this document?
    • Spencer: this looks to be in fine shape for the wg to work on
    • Jake in chat: +1
    • Poll result 10-0 for adoption, of 28 participants. Chairs will take it to list for confirmation.
    • Poll "willing to review and contribute text?": 9-1 (Eric: reason for not==no time/no expertise)

MOPS onwards [15min]

• Milestones — revisit and review (see mailing list)
  • Pushing SVA draft milestone to July, pushing SMPTE draft to November.
  • Eric: don't worry too much about the November deadline, this looks like tentative successful working group
  • Warren: +1, this working group is looking successful. Most likely we are OK with milestones later than our current committed charter.
  • Leslie: Milestone updates will be sent to list. Will aim to keep opscon at July target even if we don't progress as much.
  • Spencer: will organize an interim to target last call in July.

AoB [5min]