Media Pipeline TF call

15 Dec 2011

See also: IRC log


glenn, Bob_Lund, Kazuyuki, David, Jan_Lindquist, Russell, Clarke, Franck, Duncan, Dave_Mays, Mark_Watson, Juhani, Jason, Mark_Vickers


<Clarke> -> http://www.w3.org/2011/webtv/wiki/MPTF/Agenda_Telco_15th_December_2011 Agenda

Proposals for HTML group

clarke: prepare anything to submit for html5
... (1) network errors proposal
... (2) method #1 for ADB params, feedback

network errors proposal

<Clarke> Network errors: http://www.w3.org/2011/webtv/wiki/MPTF/HTML_Error_codes

clarke: does anyone have comments? questions?
... none heard
... recommends approval for submission

jason: are mark's parameters about levels, etc., on that page?

clarke: that will be in ADB discussion

RESOLUTION: will forward network errors proposal to HTML WG

adaptive bitrate (ADB)

<Clarke> minimal control: http://www.w3.org/2011/webtv/wiki/MPTF/ADR_Minimal_Control_Model_Proposal

clarke: comments from mark w to start
... didn't see comments on UCs
... talked about 3 models
... minimal control, limited control, full control
... minimal control is basically UA based
... full control is under JS control, including video segments to JS
... still have open issues, needs further discussion and consensus
... if we get minimal control into html5, it gets our foot in door
... basic approach to start with (minimal|limited) control in html5, then progress to more full control features
... any concerns about UCs?

kaz: wondering about media fragments
... do we want to use/support media fragment access in ADB?

clarke: is this search?

kaz: yes: (1) using time, (2) using identifier

clarke: do you want to add a use case?

kaz: wondering about this, queries if there is interest

<franck> http://www.w3.org/2008/WebVideo/Fragments/WD-media-fragments-spec/

bob: MF provides ability to access portions of tracks using # notation, similar to jason (query) notation

<mark> http://www.w3.org/2008/WebVideo/Fragments/WD-media-fragments-spec/

bob: what is relation between HTML track model and ADB?
... may have requirement for MFs, but more work needed

kaz: MF is a separate spec; we can look at this later

clarke: we may want to add as UC

bob is noisy

clarke: let's look at more as we pursue method 3
... any other comments on UCs?

jason: do want to prioritize UCs?
... need to group
... reporting, statistics, ...
... what is more important, control or reporting?
... requires discussion about prioritization

mark: should prioritize reporting
... per-user manifest vs larger scope on manifest

i didn't follow that last very well

clarke: does anyone disagree with assigning priority to reporting?

david: to group UCs

jan: separate point: measurements vs errors

<kaz> ACTION: mays to group UCs [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action01]

<trackbot> Created ACTION-85 - Group UCs [on David Mays - due 2011-12-22].

mark: what does performing "best" mean?

clarke: reporting non-subjective criteria requires judgment

mark: what does it mean for a particular bitrate to be performing "well"?
... can look at how system is performing

jason: may be condensed to single UC
... may tailor bitrates

<JanL> ?q

<BobLund> Here is the URL for the media fragments spec that we were discussing a few moments ago http://www.w3.org/TR/media-frags/

<JanL> +q

jason: as long as metrics are exposed,
... doesn't expect UA to grade

jan: suggests moving to APP layer

dave: thought that was goal???

<BobLund> Kaz - yes I rejoined on SIP

clarke: does second to last bullet item provide that?

davidmays: what is reporting to?

davidmays: requiring exposure of metrics

<mav> that's dave mays speaking

davidmays: video bit rate, dropped frames, etc

not capturing...

Clarke: UC aren't REQs
... 2 different uses regarding stats


clarke: {max,min}level, startlevel

mark: as soon as we talk about levels, the script needs to understand manifest, etc
... more in model 2/3
... model 1: independent of adaptive streaming

<JanL> +q

mark: more looking at hints or string on max/min BW
... if minlevel satisfied, should avoid rebuffering
... if not, then may move to lower quality

bob: agree with mark in general, but using text track interface, may make manifest available to scripts
... easy to specify rules to provide manifest
... manifest contains track/program description
... UA detects manifest change, then cues via text track

mark: would format be independent?

bob: not sure necessary; manifest files self idenifying

mark: would script require support for all manifest formats?

bob: could choose, but no harm done of not recognized
... may have both models

mark: easy define methods for passing info (manifest), but needs to be clear on functions
... whatever comp is responsible for bitrate adjustment, needs to operate at level of understanding bytes/frames
... knowledge at level of streams may not work

clarke: fundamental question
... abstract level vs specific/concrete
... may wish to optimize on resolution instead of bitrate

<JanL> +q

clarke: might want to do something with BW that is generic to different ADB algorithms

mark: if max BW spec, doesn't imply APP won't choose higher BW

jan: example interesting; but one concern
... how can APP process?
... must download video first
... if VBR, may go down (in rate?)
... maxLevel is to cap, in order to obtain unified experience
... see ML re BW discussion
... wants to reserve option to associate with manifest

mark: need to be clear on arch
... should look more at model 3
... on demand streams benefit from manifest data more
... within UCs, did not have objective that script makes quality uniform
... most important UC, is limiting overall BW
... esp those users with data caps

jan: wants to cover both cases (live and on-demand) where in former case, manifest is less relevant, but in latter, is more important

mark: most important to make semantics clear for script
... UA impl responsible for staying within BW limits
... hard for script to enforce

jan: may have slightly different goals (with mark)
... may leave a little bit open, further discussion over time, esp with WGs

mark: anything proposed to HTML WG won't be final
... as they will want to discuss
... must explain what we propose, the more we propose, the more we have to do
... if we tie into manifests, then will need to say much more

jan: should we change name or propose more text?

mark: would remove (?) ...

clarke: q is backed up

kaz: wondering about timing
... html5 driving timing
... may aim for next version

david: maybe max BW is separate intent from other params
... min/max to be bound to particular representation within tracks
... should we have additional param for max BW... what units of measurement?
... tricky for min/max to be bound to particular manifest format
... may be optimizations for start up, qos, ...

mark: feels like model 2 is trying to creep in
... model 2 is hard
... requires more work to flush out script control
... model 1 - only UA has knowledge of different levels
... UCs listed here is one step removed from script control

david: not control, but hints

Clarke: hints, suggestions, otherwise requires more knowledge
... level requires more definition

mav: (1) do we think that choice of BW max could apply outside of ADB?, e.g., for resolution at start?
... lost audio ...
... would like to see example from other formats, e.g., flash, silverlight
... wants to see examples from existing practice

clarke: anyone willing to comment on flash, silverlight?

jason: may provide input on flash

david: may provide input on silverlight

<trackbot> Sorry, couldn't find user - to

<scribe> ACTION: jason to provide input on practices with flash [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action03]

<trackbot> Created ACTION-86 - Provide input on practices with flash [on Jason Lewis - due 2011-12-22].

<scribe> ACTION: davidmays to provide input on practices with silverlight [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action04]

<trackbot> Created ACTION-87 - Provide input on practices with silverlight [on David Mays - due 2011-12-22].

mav: wants to see more verbiage on this practice

clarke: in last few mins, can we talk about starting level?
... can we drop starting level?
... just use {min,max} as proxy?

mark: are we in model 2 or model 1?
... doesn't see way to talk about levels in model 1

<Juhani> See for different systems: http://www.w3.org/wiki/HTML/next#Adaptive_Streaming

mark: confusing models a little bit
... if we're doing model 2, then comparison with flash/silverlight more relevant
... in model 1, only can give hints on what user prefers
... not convinced about providing such choices to users
... but content providers might want this

<Juhani> Sorry, wrong reference ... right one is http://wiki.whatwg.org/wiki/Video_Metrics if applies to the discussion earlier of how different systems use parameters

notes we are now out of time

clarke: continue discussion on reflector

<Clarke> Thanks, Glenn

<kaz> thanks a lot, Glenn!

Summary of Action Items

[NEW] ACTION: davidmays to provide input on practices with silverlight [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action04]
[NEW] ACTION: jason to provide input on practices with flash [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action03]
[NEW] ACTION: mays to group UCs [recorded in http://www.w3.org/2011/12/15-webtv-minutes.html#action01]
[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.136 (CVS log)
$Date: 2011/12/16 18:20:01 $