Interacting with the Networking Layer -- 28 Oct 2015

dom: could get better perf if better interaction between network and application layers
... you also want more resilient interaction with the network
... tensions between neutral network and will for performance
... tensions between neutral network and will for performance
... how can W3C help bridging that GAP between the 2 layers
... main goal of this session is to have an open discussion to see if there is something W3C should do in this area

Dominique Hazael-Massieux: introduction to the needs and constraints of getting apps and networks to work together

Report from the recent MaRNEW workshop

Natasha Rooney: report from the recent MaRNEW workshop and experience of networking Task Force in Web & Mobile IG

-> https://www.iab.org/activities/workshops/marnew/ MaRNEW workshop

natasha: 2 major reasons that pose issues to operators
... management of network
... and politic area (regulation on censorship or legal interceptions)
... at the workshop first session was around encryption
... next session talked about IETF work on deployment considerations
... ubiquitous encryption
... propose some way to operators to improve without breaking encryption
... another session was and trust models and user choice presented by Wendy
... good statistics
... session 3 sending data up for network management benefits
... session 4 is sending data down for network management benefits
... network saying here is the current situation
... TCP was mentioned a lot we had a session dedicated to the transport layer
... session on application layer optimization
... we needed an overview from content providers
... session 6 was on transport layer
... congestion control innovations
... output is a list of ideas
... willl be at IETF next week
... evolving TCP
... mobile throughput guidance
... one bit latency bw tradeoff
... (to see how many bits we need to classify traffic)
... blind caching
... encrypted content
... metrics for net statistics
... and also better collaboration as we have been working in silo for a long time

mark: interesting idea was trust verifying
... benefit of optimzation should be served between the network and content providers
... second thing was about qos if you can trade latency
... so a time varying thing

How is the network venue

tatsu NTT Communications

tatsu we are providing WebRTC on robots

scribe: have a demo booth

tatsu one option is to provide a hierarchy of model

tatsu webrtc P2P traffic, P2P doe not mean it is best approach, to provide better service for the user have to find best path

tatsu collaboration with web application and SDN

scribe: on W3C web things and network things best architecture should be achieved thanks to statistics
... capability for selecting best path

natasha: I like the skyway project

yoav: have you made any contribution to webrtc as part of this project ?

natasha: we started working on data plan information API
... you can have info on usage of data
... how much traffic is being used if dedicated to a particular application, info to users how much they have left on their data monthly allowance
... Google-Fi network display this quite nicely
... providers are keen to work on this but their operations team don't
... info good in some countries
... how could operators not leak too much info and still keep the info relevant
... GSMA is interested in this again
... tell me if you are interested

frode: I'm an operator we don't want to give this information away
... it is critical business information

yoav: it would be great if it would be obfuscate that data and make it available

mark: technical solutions that take human out of the loop are the way to go

cullen: most people with a 3GB plan are actually below that

yoav: when peopel are reaching their limit or if they have roaming cost the application could limit its data usage

jorg: I'm from DT and come from Identity space, we need to give control to the user so that he can make an informed decision, it becomes compelling if user can buy the extra bw to use a given service
... network for dedicate wishes the user is the customer so needs to be in the middle of the decision
... currently we are still trying to meant things trying to optimize things when we can't

dom: having a 3 party negotiation between the network the service and the user

mark: our experience is that a solution that depends on teh user will liklely not work, we have an option to adjust bw usage in Netflix but users are not using it

yoav: app developpers are making not always assumption, like are you on wifi ? meaning network is free which is not always true
... need a communication channel for these metrics, what type of network communicate, bw information even if not accurate, communicate cost
... we need to get rid of assumptions and align it with metrics that are meaningful for the user

frode: a solution could be to have this info inlined

MarkF: multiple applications seems there is a rule for the network to allocate resources

dom: one of the needed optimization, you want to avoid waking up the radio layer to save battery, I think you optimized it at Google

igrigorik: talking to the driver to get more detailed information, eg quality of the signal.

dom: app developpers should understand how to organize their data usage

igrigorik: an example is the search bar, it was slow when sending a small amount of data, solution was actually to send more data with empty data to make things faster

dom: is that something operators could expose ?

s/igrigorik/s/alia/igrigorik//

igrigorik: sometimes you have to prioritize some data even if it is a small amount of data

cullen: we know how to optimize apps need to do it for the web

martin: how is it different ?

mnot: we discussed this in HTTP world
... we need some way to distinguish this on the wire especially if it is encrypted

cullen: from an API point of view I wish we could do the optimization

igrigorik: we have some info in the browser to do some optimization

dom: are you saying the app developers don't need to know this ?

igrigorik: maybe it could be exposed

speaker6: could be dangerous everybody would say its high

dom: the browser could mediate this
... fetch as a priority

igrigorik: currently its only a placeholder
... I am curious of optimizations done by native apps

martin: webrtc can prioritize streams 4 different options
... not yet implemented
... not worky in firefox for sure
... we take patches

cullen: native communication apps all tries to do optimizations

igrigorik: if those APIs exists browsers should be leveraging them

mnot: there i a new TCP Performance Tuning for HTTP draft from Daniel Stenberg on this topic

jorg: sounds we have a few approaches on solutions

<mnot> FYI, Daniel's draft: http://bagder.github.io/I-D/httpbis-tcp/

jorg: is it possible to have one abstraction for network providers to expose their capabilities

varun: @@@v

jorg: who could define what would be the requirements? thinking W3C should be the right place for this.

frode: I can almost guarantee operators will not adapt traffic based on this marks

cullen: I think you need to know how to trust this information, sample and trust is important

dom: close to the end of this session, do we have clear follow up we should identify

MarkF: @@
... also what should be exposed to developers

dom: is the tradeoff between latency and bandwith part of this ?

natasha: maybe we need to go ahead and do this MARnew workshop

dom: if you have ideas please come to me
... last year we created a TF, if people wants to contribute and revive it
... thank you all for attending

- DRAFT -

Interacting with the Networking Layer

28 Oct 2015

Attendees

Contents

Dominique Hazael-Massieux: introduction to the needs and constraints of getting apps and networks to work together

Report from the recent MaRNEW workshop

How is the network venue

Summary of Action Items

Scribe.perl diagnostic output