Sudeep: [reviewing agenda]
... we will present the TPAC agenda on which we're seeking feedback
... we will then discuss a use case, with a particular illustration from an Intel team, who will also give a more in-depth presentation during TPAC
Dan: TPAC is very close - it's important to focus on the topics we're trying to cover
... we will be meeting on Tuesday from 8:30am to 5pm
... we would start at 8:30 with logistics, intros, some background and concrete references that Dom will be presenting
... then we'll have a break - always great opportunity for informal discussions during TPAC
... then we'll dive into guiding principles for the "hint framework" - going into examples and spending some time analyzing what worked and doesn't
... then Song will cover an update on liaisons
... then lunch, after which we'll get our guest speaker session with an Intel team presenting a more detailed explanation of the solutions they've been working on
... that could form the basis for an API
... after the afternoon break, we have more use cases review - we're looking for input and feedback from participants on how to drive this effectively
... we will have many opportunities to touch upon use cases throughout the day, but we want to take advantage of the F2F to go into more details in these use cases
... we will have IG members, observers - it's a good opportunity for a dialog
<cpn> scribenick: cpn
Dom: We have 50 observers registered, which is a good sign, but unlikely they'll all come. We'd need to take into account the size of the group
Dan: We can use sticky notes to
gather ideas for discussion. Then we can combine the ideas, or
work from the prepared agenda. We should be open to suggestions
from members and observers.
... It's important to have a bi-directional communication, not
just have presentations
<dom> scribenick: dom
cpn: in the Media &
Entertainment IG, we've dedicated to a "look-ahead" session for
new use cases & requirements
... one of the tools that we will be using is a flipchart to
keep track of ideas as they emerge
... inviting people to share their thoughts on unaddressed use
cases
... it is a challenge - not everybody has something to
contribute, many people will be observing rather than
contributing
... in that situation, it's up to the chairs of the group to
really to come with your own ideas to seed the conversation
Dan: agree that we'll need to
stimulate the dialog
... the idea of having sticky notes to keep track of ideas as
they come up would be useful
... even if we can't review them all
... we'll have to make sure we have the right logistics
... I really want this to be a two-way communications
... although I expect it will be hard to break the ice
Sudeep: assuming we identify new
requirements out of this session
... would it make sense to schedule a follow-on breakout
session to look at how to addressing that requirement
... q-
... to get feedback on what we would have identified
cpn: I would strongly recommend
that you do that
... I hope that the session grid will have space
... we've done that often with the Media & Entertainment
IG
Dan: For the Wednesday session,
the door is open for ideas
... this year the suggestion for sessions are collected ahead
of time
... we could place a stakeholder
... Getting back to the agenda, I think it's worth injecting a
5-10 minutes early in the morning to explain this process
Sudeep: we will have remote participation enabled for TPAC if you can't make it to the meeting
DanD: on another note: lunch is a good opportunity for informal discussions
[Presenting slides: Status in liaison organizations and topics for TPAC]
Song: first, some background on
existing activity in W3C
... Network information API is defined in the WICG
... last updated last February
... The API enables Web apps to access info about the network
status (e.g. wifi, wirless, bluetooth connection)
... Second, WebXR - discussed in the Immersive Web WG
... pretty active ongoing work
... WebRTC is one of the most important WG in the org, nearly
done
... Web of Things Architecture, in CR, last modified in May -
IoT has strong links to networking
... In terms of candidate items for TPAC
... 5G Network Slicing - network slicing divides networks in
multiple virtual networks based on app requirements
... we're trying to establish a liaison with the relevant
group
... Edge computing enables integrating network, computing,
storage in nodes
... operators have developed their own edge computing
APIs
... China Mobile has a Web MEC API proposal with several
scenarios
... (face detection, WoT, Edge caching for UHD service)
... it works with MECNodeService to which one can register and
then send requests
... Because Edge Computing is pretty important for apps in new
networks (e.g. 5G), some operators have defined their own APIs
already
... we're looking at liaisons to ETSI & 3GPP who have
defined set of APIs (incl location, bandwidth management, radio
network)
... Real-time network adoption is another interesting area of
interest
... based on mobile throughput guidance information
... Cloud gaming needs further thinking
... Best network selection - how to determine which link to use
- related to network slicing
dom: will you give a more in-depth presentation on Web MEC?
Song: this is still a very early
stage proposal
... we're still working on modifying it - once we're done, we
will be able to share more
Picky: on your slide on MEC - how does this intersect with 5G data plans
Song: let's revisit that offline
Network Link Performance Prediction
Sudeep: Jonas and Jon are IG
participants from Intel
... they've been working on topics that are overlapping with
some of our discussions esp around hints
... they'll give us some introduction to their work to
stimulate our discussions
Jonas: we will be in TPAC
... The work was stimulated by 5G
... 5G brings better performance, but also more variation
within networks
... as we're moving towards edge compute, this creates also
more variation
... edge can mean many different things which induce many
different behaviors
... Networks are "best effort" today - we have been interested
in seeing how we can make them more deterministic and to better
inform app of their network environment
... We've been investigating network link performance
prediction - not in a Web-specific context
... one starting point is that we don't want to control apps -
hence the "hints" approach that matched what has been discussed
in the IG
... we're looking both at current and future link
performance
... along a number of parameters: bandwidth, latency, cell load
and more
... a typical example where this is relevant: if you're driving
your car, you may going between a macro cell with low load then
high load and back to low
... a network-aware app could pre-fetch data to take these
expected changes into account
... e.g. in a video streaming context
... the same could be applied to coverage gaps
... We've also looking at the challenge of mmWave which have
very small coverage and very high throughput
... there are other cases where this applies, but these are the
key areas
... We're making these links prediction as a service on the
side
... it's transparent from a client/service perspective,
available to both
... the point here is that it should be agnostic to cloud or
edge
... we're also not touching the data streams itself
... it would typically deployed in the operators network and
then fetched from client or server
... in TPAC we will dive in more details in terms of the
predictions are generated, and how they could be exposed in Web
apps, in browsers, in servers
... We have a number of use cases that illustrate how the
predictions could be used
... A simple case is media streaming - you need to look at the
value add across the various network stakeholders
... There is a benefit here to the end user, to the content
provider and to the operator
... in media streaming, this means less buffer delays for
real-time media
... for non-real time streams, it helps reduce the amount of
data transmission in the network
... we will cover these additional use cases in TPAC as
well
... We're very interested in getting feedback and reactions
dom: very exciting! looking forward to hear more at TPAC
sudeep: note relationship to the "smart offline download" use case
dom: agreed - the notion of "gap coverage" can be well related to the notion of cost/cell load
jonas: we're looking at usage behaviors to inform our research on utilizing this
sudeep: there will also be a demo during TPAC?
jonas: one keep example is media
streaming; next will be cloud gaming which is more driven by
server-side prediction
... we will be present a cloud-gaming set up to show the
benefits of link prediction
... we will show also media streaming with and without link
prediction
cpn: where will the demos be?
sudeep: there is demo slot on Wednesday
cpn: I'll mention it to the Media IG
Sudeep: the slides that were
presented are available on github - the discussion can continue
there
... we will continue this discussion at TPAC
... we will share the Webex details for the meeting there