Minutes IETF105: mops
minutes-105-mops-00
| Meeting Minutes | Media OPerationS (mops) WG | |
|---|---|---|
| Title | Minutes IETF105: mops | |
| State | Active | |
| Other versions | plain text | |
| Last updated | 2019-08-13 |
minutes-105-mops-00
Minute takers:
Kyle Rose
MOPS (Media Operations BoF)
IETF 105 Montreal
7/22/2019 1:30 pm Montreal time
Leslie Daigle introduces the BoF
* Gave a background of the problem
* Goal is to launch a WG or other formal group for regular meetings.
SMPTE 2110 Suite - Digital Video on IP Networks
Presenter: Glenn Deen
* draft-deen-mops-smtpe-2110-normative-references-00
* Summarized the transition to IP on the part of professional video outfits,
and what kinds of IETF-standardized technologies they will need.
Video Interest Group side meetings at IETF 98-104
Presenter: Glenn Deen
* History of the first VIG BoF at IETF 98, and the following side meetings
* Demonstrated diversity of organizational attendance at side meetings
Open Caching: Enabler for OTT Streaming Video
Presenter: Sanjay Mishra
* SVA wants to address OTT streaming video at scale
* Open caching initiative based on CDNI RFCs
* Opportunities for progress across httpbis, TLS, ACME, CDNI, others
* Issue with where to go to beyond CDNI RFCs (8008, 8007)
* Challenges include different interpretations of HTTP redirect semantics,
bitrate shifts/rexmits, very low latency and high throughput for VR * MOPS is
to be a forum within the IETF for orchestrating this work * QA
* Cullen Jennings: Is there a liasion with SVA to assist in this work?
* Glenn (chair): Want to provide a forum for "soft landing" of
SVA folks as they're trying to figure out what they need
* Emile Stephan, Orange: Opportunities for MOPs charter? Slide 4
* Answer: Opportunity for SVA to reuse IETF work.
* HTTP WG – RFC is clear but interpretations may
differ. Perhaps guidelines for interpretation by SVA to
bring back to IETF RFC 3986 WG.
Low Latency Video Streaming
Presenter: Justin Uberti
* Highlights real(first?)-world problems with differences in live video
streaming latency * Majority of all traffic on internet today is video * Deep
client-side buffer to avoid underflow, leading to high latency * Interactive
streaming video requires sub-second latency (Twitch, Stadia, etc.)
* Smart streaming (i.e. no CDN, little or no playout buffer.
* Existing CDN tech does not handle well.
* Stadia
* Designed to match 150 ms E2E (e.g., time from hitting a button to
seeing the character jump) * 1080p 25Mb/s; 4320p (8K) 200 Mb/s
* Low-latency interactive stack is complicated with little off-the-shelf
* Can the stack be simplfied with QUIC?
* QA
* Bob Briscoe: What about impact from other traffic? Mentions work on
L4S. * Roni Even: Why do bitrates here don't look like those from
Netflix?
* Justin: Netflix optimizes each chunk for VOD; for gaming, can
only encode once using hardware to achieve low latency, and low
tolerance for artifacts, so bitrates are higher for a given
resolution
* Roberto Peon: Compressed video is inherently stateful, meaning you
need other state to be able to decompress, which means lower latency
requires less inter-state dependence, which means there is a tradeoff
between bits-per-pixel and latency
Media Issues Taxonomy
Presenter: Jake Holland
* draft-jholland-mops-taxonomy-00
* Intended to be a quick and painless orientation for video streaming issues
* Includes an analysis of why unicast video streaming doesn't scale: Akamai's
record for 4K delivery equates to the 179th most popular broadcast show in
terms of viewers * Congestion avoidance via ABR, feedback-driven streaming *
Video conferencing requirements and options * Implementations: should MOPS
propose maintaining such a thing within the IETF? * QA
* Cullen Jennings: Does this doc have a comprehensive list of the
problems faced?
* Jake: Want someone closer to the problem space to write a
roadmap of all the problems and proposed solutions * Leslie
(chair): We've already seen in this meeting that there are
different kinds of video: "video" is not one thing * Jake:
Seems like we need a forum for posing problems/questions and
proposing solutions/enhancements (e.g., to HLS)
* Sanjay: Specific areas you want to focus on?
* Jake: Want collaboration so we can identify a more-or-less
complete set of problems faced by the video streaming industry,
as well as identifying areas for potential research
* Tony Tauber: Would be nice to get some standard numbers for some of
these problem areas so we can design networks that meet expected
requirements for use cases * Kesavan Thiruvenkatasamy: <Identified two
issues, didn't catch them>
Scoping Discussion
Presenter: Leslie Daigle
* Draft charter already compiled, based on MBONED and v6ops
* Draft charter QA
* Glenn Deen (chair): IETF wants to know: what problems are we trying
to solve? One recurring problem is that putting IETF-designed pieces
together often runs into roadblocks around boundary cases. Hard to find
a place to address these issues with an audience of experts. One way
MOPS can be useful would be to look at problems encountered by video
users and delegate technical work to appropriate areas of the IETF. *
Darshak(?): <didn't catch that or Leslie's response> * Justin Uberti:
This draft charter implies that this is targeted more toward operators
than toward technologies. Is that right?
* Leslie: Not the intent
* Matt Stock: Value is being able to answer whether something is a good
way to solve a problem, and to be a source of best practices.
* Leslie: Not just about inefficiencies. Some current users of
video are challenging some of the assumptions made in network
topology and scaling and in the design of protocols. * Glenn
(chair): History of evolution of video streaming, but
highlights there is a long way to go: right now, it stresses
the protocols and networks.
* Stephan Wenger: Suggest to focus on mass distribution of video,
versus P2P video conferencing. Some convergence between those things,
but sacope seems too broad.
* Leslie: Not sure that's the right way to tighten, but it's a
useful point.
* Roberto Peon: Similar fears: this is very broad. For this to be
successful, we need to have a really crystal-clear idea of what's
in-scope, partly not to overlap with work being done elsewhere.
* Leslie: We want to take work that overlaps with other areas,
and dispatch it to them.
* Goals QA
* Jason Thibeault: Figure out a way for the IETF and SVA to work
together, without the IETF trying to co-opt the work the SVA is doing.
* Cullen Jennings: How do we expect MOPS-developed BCPs to constrain
work done elsewhere?
* Leslie: Way to document how operators should handle certain
kinds of traffic. * CJ: Need to make sure BCPs are actually
best and current. Can't think of anything that isn't in-scope
for this charter.
* Ali Begen: (via Jabber) "I wonder how many folks are in the room who
are not regularly IETFers. To make this WG a success, working closely
and together with bunch of several other organizations is a must."
* Glenn Deen (chair): Hard to find a place to do the kind of
work that MOPS is proposing.
* Alissa Cooper: Mixes a few useful things. Usually a charter would be
a little more specific about actual work items. Separate into "here's
the work list" and "here are our loftier goals". Also, decide which
things need WG consensus and which don't. Also figure out which
protocols we're extending. * Roni Even: What does "user" mean in #1?
For #6, whch other forums are relevant for cooperation. * Stephan
Wenger: Don't include acquisition, as it's a fundamentally different
problem.
* Glenn Deen (chair): Two things go hand-in-hand.
* Warren Kumari: Can we get a show of hands of who is not a regular
participant in the IETF?
* Leslie called for hands. The number of people who are not
regular participants who only came for this BoF was nearly zero.
* Roberto Peon: Media "acquisition" is or is not the same thing as
ingestion...? What is the intent?
* Glenn Deen (chair): Just ingest/getting the video into the
system.
* Charter scoping paragraphs QA
* Chris Lemmons: The "ship it off to another WG" is interesting. When
you try to hook together a bunch of protocols, you might end up needing
extensions, so we need to make sure to send that work to the right
places.
* QA
* Lenny Giuliano: Not obvious from the text of the overall draft what
exactly the problem is that you're trying to solve. But a coordinating
role by a WG like this would be very valuable. MBONED would like to
know if the work they're doing would be useful to the participants in a
group like this. * Leslie: The IETF doesn't really have a structure
appropriate for the VIG, so we're trying to shape this in a way that
works within the confines of the IETF, which is part of the reason why
it seems like a grab bag of stuff. * Jake Holland: Would it satisfy
breadth-of-scope concerns of those if we limit it to solving
video-related problems?
* Roberto Peon: Gaming has almost exactly the same concerns as
video, so the suite of technologies we develop will be very
similar, if not the same.
RFC 5434 Questions
* Hum on whether there is a problem here to be solved, and that should
be solved within the IETF.
* Hum solidly in favor, none opposed
* Cullen: Concerned that people can't articulate what the
problem is.
* Who is willing to participate in mailing lists and write/review
drafts?
* Close to 20 hands
* Is the scope of the problem well defined? (Laughter.)
* Lenny: Is there non-video media the group is looking to
coordinate?
* Leslie: Are games video?
* Lenny: If it has video, then it's video.
* Leslie: Not sure the distinction is helpful.
* Stephan: Problem is multi-multi-multi-megabit
streams, not small audio streams. * Glenn (chair):
Video *is* the media because video is often used as
shorthand for video+audio+whatever else. * Justin:
There are other high-throughput live streaming cases
that don't qualify as "video" that might be appropriate
for MOPS. * Cullen: Wouldn't rate inclusion by
bandwidth only. * Jake: Not just about bandwidth, but
also about realtime/latency requirements. * Lenny:
Concept of "mass" * Harald: Think about defining a term
that means something that at least contains a video
component, and has some kind of timing (deadline)
constraints on delivery. Then, use that term throughout
the rest of the charter. * Sanjay: What about
measurements? * Chris Lemmons: Might be most useful if
we think clearly about specific things that are *out*
of scope. * Roberto: Implied are stateful things,
consumed in real time, session based.