- From: Dominique Hazael-Massieux <dom@w3.org>
- Date: Fri, 20 Mar 2020 15:40:39 +0100
- To: "'public-networks-ig@w3.org'" <public-networks-ig@w3.org>
- Cc: "Sabella, Dario" <dario.sabella@intel.com>
Hi,
The minutes of our teleconference earlier today are available at:
https://www.w3.org/2020/03/20-web-networks-minutes.html
with Dario's slides at:
https://www.w3.org/2020/03/20-web-networks-5Gaa-predictive-qos.pdf
The text version of the minutes are copied below.
Dom
5GAA Edge Computing and Predictive QoS - Web & Networks Interest Group
Meeting
20 March 2020
[2]Agenda. [3]IRC log.
[2]
https://lists.w3.org/Archives/Public/public-networks-ig/2020Mar/0001.html
[3] https://www.w3.org/2020/03/20-web-networks-irc
Attendees
Present
Chris_Needham, Dan_Druta, Dom, Eric_Siow, Huaqi,
Jon_Devlin, Jonas_Svennebring, Louay_Bassbouss,
Peipei_Guo, Piers_O_Hanlon, Song, Songfeng_Li, Sudeep,
Taki_Kamiya, Xiaowei_Jiang, Xiuquan_Qiao, YKhuang
Regrets
-
Chair
-
Scribe
dom
Contents
1. [4]Intro
2. [5]Presentation from Dario Sabella - MEC in ACTION: An
Overview of Edge Computing Activities, Part2: Predictive
QoS for Edge Computing, insights from 5GAA
Meeting minutes
Intro
[6]Intro slides
[6] http://www.w3.org/2020/03/20-web-networks-intro.pdf
Sudeep: welcome to all - this is our 8th teleconference meeting
- hope all is doing fine in this difficult times, thank you for
taking the time to join this call
Sudeep: today's agenda will cover 2 call to actions, followed
by a presentation on 5GAA from Dario
Sudeep: 2 call for actions:
… the network information API, a work item driven by the Web
Platform Incubator Community Group, is a candidate for
inclusion in a WG charter
… we had an overview of the API a few weeks ago, and its
implementations in Chromium
[7]Call for any requirements for NetInfo API
[7]
https://lists.w3.org/Archives/Public/public-networks-ig/2020Mar/0002.html
[8]Network Information status?
[8] https://github.com/w3c/dap-charter/issues/78
Sudeep: please have a look and see if you have input on this
proposal
… Another call for actions: a request from a W3C Member
company, Peer5
… they've started a discussion around requirements for P2P CDNs
with host-to-host connectivity, around IP address obfuscation
[9]Local IP obfuscation #59
[9] https://github.com/w3c/webrtc-nv-use-cases/issues/59
Sudeep: this is also relevant to the scope of our IG - we've
had presentations around P2P CDN before
… please take a look at the problem statements on the github
issue
… and provide your input
… Upcoming discussion in the IG will include a presentation
from Cloudflare that Dom is arranging
… Now, handing over to Dario to talk about some of the insights
from the 5GAA on predictive QoS
Presentation from Dario Sabella - MEC in ACTION: An Overview of Edge
Computing Activities, Part2: Predictive QoS for Edge Computing,
insights from 5GAA
[10]Presentation slides
[10] https://www.w3.org/2020/03/20-web-networks-5Gaa-predictive-qos.pdf
Dario: this is a follow up to my presentation in January when
we ran out of time
[11]Minutes of the previous presentation
[11] https://www.w3.org/2020/01/15-web-networks-minutes.html#item02
Dario: I'm an Intel delegate for 5GAA, also vice-chair of MEC
… last time we covered an intro to edge computing, the MEC
standard and the use cases of interest
… Today we will cover Predictive QoS in the context of 5GAA
… 5GAA is a huge association, with over 130 members coming
mainly from 2 family of stakeholders: automotive industry and
telecommunications
… The goal of 5GAA is to bring together automotive, technology
& telecom comapnies to work together to develop e2e
connectivity solutions for future mobility and transportation
services
… the goal is to make two worlds together, across a set of
skills, perspectives, languages
… I'm coming from the telco background and it has been very
useful to listen to requirements from car makers and turn those
in terms of telco requirements
… it includes many aspects of C-V2X incl technology,
standards... incl business models
… [5GAA Org structure]
… There are a number of WGs dealing with different activities
(use cases & requirements, architecture, testing & pilots,
standards & spectrum, business models, regulatory, security &
privacy)
… [slide: What is C-V2X?]
… C-V2X: "Cellular-Vehicle to Everything"
… this includes vehicle-to-vehicle (V2V), vehicle to pedestrian
(V2P), vehicle to infrastructure (V2I), vehicle to (cellular)
network (V2N)
… bringing network to the car brings another "sensor" to the
car
… it enables communications with other cars, other (vulnerable)
road users such as cars and bikes, and infrastructure
… the 5GAA is collaborating with many organizations
… Looking more specifically at Edge Computing / MEC - it is a
key technology for many use cases in 5GAA
… [slide: Automated and Connected cars - key drivers]
… there are 10s or 100s use cases defined by automotive
stakeholders in detailed way, with a number of white papers
published
… see www.5gaa.org
… One white paper might be of particular interest
[12]Toward fully connected vehicles: Edge computing for
advanced automotive communications
[12]
https://5gaa.org/wp-content/uploads/2017/12/5GAA_T-170219-whitepaper-EdgeComputing_5GAA.pdf
Dario: more recently, we've been looking at Predictive QoS
… this is a hot topic not only in 5GAA, but also in ETSI MEC,
3GPP
… this is seen as a key mechanism
… the idea is to enable a notification in advance on an
upcoming network quality change (either better or worse)
… which enable to react depending on the use case
… [Slide: Predictive QoS in 5GAA]
… depending on the use case, the KPI to predict will change,
with different type of reactions at the application level
… for instance, for a tele-operated driving, reaction might
include changing route, parking the vehicles, etc
… to convey this information, the idea would be to reuse some
of the existing APIs, e.g. MEC Service API
[13]Making 5G Proactive and Predictive for the Automotive
Industry
[13]
https://5gaa.org/wp-content/uploads/2020/01/5GAA_White-Paper_Proactive-and-Predictive_v04_8-Jan.-2020-003.pdf
Dario: There is a stable draft in ETSI MEC for the MEC V2X API
(MEC 030)
… it should be finalized in the upcoming few weeks and
published as a standard soon
[14]Multi-access Edge Computing (MEC); V2X Information Service
API
[14]
https://portal.etsi.org/webapp/WorkProgram/Report_WorkItem.asp?wki_id=54416
Dario: this is just one functionality - stakeholders are free
to define their own APIs as long they following the same REST
patterns
… MEC isn't trying to define all the possible APIs of the 1000s
of potential services
… I would like to propose to focus on predictive QoS based on
the "infotainment use case", for which the KPI to be predicted
is the data rate
… where the expected reaction would be to adapt the video
quality
… the idea is to imagine a streaming video content - this is
similar to the traditional video streaming, with the addition
of mobility requirements
… The experiment we ran showed the impact of predictive QoS on
this use case
[15]5GAA live demos show C-V2X as a market reality
[15]
https://5gaa.org/news/5gaa-live-demos-show-c-v2x-as-a-market-reality/
Dario: [Slide: In-Vehicle entertainment utlizing MEC over 5G
Networks]
… The demo, led by Intel, Marelli, Terranet, Equinix, showed
how MEC can support immersive HD entertainment for all
occupants
… incl video streaming, gaming, VR, office work, online
education, advertisement
… the underlying assumption is that the car is using the
network and has access to edge computing capabilities
… This use case matches a promising market with predictions of
$7 trillion passengers economy - with 300 hours a year spent
behind a well, lots of opportunities
… [Slide: predictive QoS - reference scenario]
… Imagine a car going from one point to another; the car has a
client app, which connects to an edge node
… there is a model of network characteristics mapped over the
path
… MEC is used to convey these characteristics to the car
… the said info can be used to adapt network usage
… [Slide: Predictive QoS - reference scenario]
… Different use cases would be usable in different zones based
on the QoS circumstances of the area
… This is import for car makers to enable or not specific
autonomous driving services depending on the network
… this is thus business critical for them
… We have demos that covered 2 main concepts
… one shows the benefit of edge computing with different
deployment options (e.g. how close the edge is to the user)
… the other one shows the impact of predictive QoS on
multimedia delivery adaptation in a field trial in an urban
environment
… [slide: benefits of edge deployments]
… The car app is communicating with the edge node, and a remote
cloud node
… the car app is displaying HD video content to the passenger,
as well as location-relevant information (e.g. advertizing)
… the content can be retrieved from either the edge or the
cloud - the demo looks at the various level of "edge" scenarios
… [Slide: reference architecture used for IVE App Tests]
… [Slide: MEC Performance evaluation: preliminary comparative
results]
… this shows difference in Round-Trip-Time depending on the
server locations, esp using transcontinental servers
… The second part of the demo shows the impact of predictive
QoS on adaptive streaming
… the first part of the work was to collect data on live
measurements in Turin, sampled every second during a drive test
… this was coupled with a realtime emulation done by Univ of
Pisa recreating network quality from the collected data
… The idea is to be able to make reliable predictions
… this was part of the Smart Road project for the city of Turin
… the emulator was fed with the data from the drive test
… then the MEC server was doing the video streaming adapted to
the predicted evolution of the network
… this had a huge impact on the quality - e.g. this removed
freezes
… [Slide: Predictive QoS - impact on performance, preliminary
results]
… we collected data on packet interarrival time
… the more regular the video packets, the better the reception
- better for buffering and playout
… both high interarrival time or high jitter have bad impact on
the quality of experience
… you see a big difference on average latency, but even more so
on the jitter of packet interarrival time, leading to much
fewer freezes
… This concludes my presentation
… Predictive QoS is key to automotive stakeholders, esp in a
set of well-defined use cases
Sudeep: thank you Dario for these great insights on 5GAA & QoS
EricSiow: in your view Dario, where are the collaboration
opportunities between this community and 5GAA?
… Are there clear needs for APIs where we could help bridge the
gap?
Dario: currently, MEC & Edge Computing are key strategic areas
for 5GAA
… there is a working item spanning two domains of 5GAA called
"MEC for Auto"
… I see possible synergies in this space - that work item is
diving deeply into the topic, not only in the network side, but
also for the application side
… that can be a point of contact with this group in W3C
… the work on predictive QoS is more or less done with the
white paper published
… we have to ask ourselves - differentiating the actual
prediction of the QoS from how to use the predictions
… making the prediction is probably not the most relevant for
this group (it can be done in many ways - network, app, device)
… and then how these predictions would be used is likely more
directly relevant
EricSiow: any thought how the 5GAA IPR policy might impact
collaboration with other organizations?
… In W3C, all contributions are royalty-free; in cellular
space, royalties are common
… has that been considered?
Dario: there is an IPR policy with well-defined procedures in
5GAA
… but I'm not the right person to give details on this
… the 5GAA secretariat would be the right group to approach to
figure out how this would be addressed (e.g. via a MoU)
… I'll be happy to establish the connection if/when needed
Song: with your experience Dario, from 5GAA perspective, who
will be the user for the V2X APIs? are they going to be the
developers from car manufacturers? or third-party / independent
developers?
Dario: the question is who will be developing the service - in
automotive, this is generally a heterogeneous ecosystem
… with car manufacturers, their partners
… BMW has started to act as data center of consumers that can
be used by third party / startup in a sort of app store, under
the consent of the end consumer
… any of the stakeholders can be the API consumer
Song: since MEC is quite new for W3C, I would like to borrow
some experiences from you to identify who will be the target
developers within W3C
Dario: my presentation today was focused on automotive - but to
be extra clear, MEC is not at all automotive specific
… it would be e.g. completely in scope e.g. for gaming
… for a W3C perspective, I think ideally you should look at
widening the net to more use cases (e.g. VR as well)
Sudeep: thank you Dario for this presentation - we will follow
up by email
Received on Friday, 20 March 2020 14:40:46 UTC