Meeting minutes
Slideset: https://
New charter
Sudeep: we're operating under a new charter, set to last for 2 years
[round of intros]
Sudeep: so far we covered 3 workstreams: edge coputing, network quality monitoring and emulation tools and traces
… we've seen lots of interest on edge computing and continued interest on monitoring
Sudeep: in our new charter, the scope remains mostly the same with additional focus on edge
… client-edge cloud coordination was added as a key topic, following up on our draft note about related use cases
Sudeep: in terms of emerging use cases that are driving interest: offloading capabilities, new traffic patterns with increases in uplink usage, emergence of edge networks, and impact of 6G & AI
… how will these impact the end to end user experience and what developers can accomplish ?
Sudeep: given the many intersections, we're considering to schedule regular, the Chairs would like to suggest regular monthly meetings to make progress
Eric: re hints from apps to the network and vice-versa, does it cover the ability of app & networks about hardware capabilities (e.g. on-device accelerators)
DanD: it can be as broad or narrow as we want; what's on the table at the BOF in IETF is to address a handful issues with video streaming, driven by needs from Meta and YouTube to self-police their traffic to reduce it on the wire
… because of restrictions some mobile subscribers - they want to set the adaptive bitrate to adapt to these subscription policies
… to optimize the user experience on their end
Eric: with the rise of AI workloads, getting hints on the device capabilities to the app and network sounds increasingly relevant
Cloud Edge Client Coordination
Song: during TPAC we discussed the creation of a dedicated CG
… I've been working on a draft scope for such a group with Max and Larry
Client-Edge-Cloud coordination Use Cases and Requirements
Song: the CG would start from the use cases to develop a proposal for offloading
… the CG could develop requirements, a proposed architecture for network offloading, and allows coordination with other W3C groups (Web & Networks, WebPerf, WebRTC)
… we'll publish an initial draft soon
Max: ["The mission of this group is to provide mechanism and interfaces between Central Edge, Edge Cloud and Client for computing work load offlaoding and orchestration. the typical use cases include AI generation, cloud gaming, streaming acceleration, etc. The goal is to design a set of new APIs and mechanism that enable computing workload offloading and orchestration between Central Cloud, Edge-Cloud and Client. It should leverage existing mechanisms
as much as possible and should coordinate with related W3C working groups and other SDOs and open source community if necessary]
Max: proposed name would be "cloud edge client coordination CG"
q/loa/loa/
Max: proposed short name would be "cloudedgecoord"
… comments are welcome
Dom: suggest making sure use cases and req continues in the IG
… the note has 2 architectures, would the CG focus on both or only one of them?
… the name could be improved
Song: makes sense
Chris: let's iterate on the proposal in github
Max: re which architecture in the CG, from my perspective, we should be open to any solution with more detailed research before picking one or several ones
… and +1 on iterating on the name and other aspects in github
Sudeep: an additional input: in the IG we would focus on networking factors; in the CG, would this expand to device/compute/storage capabilities?
… including discovery
Song: exposing capabilities may be difficult given the sandox and security constraints, but this could be part of what the CG can discuss
Sudeep: so let's upload this on github to iterate
Song: will do
Other topics for consideration by the IG
Dom: would it be interesting to hear more about IPVM which Robin alluded to?
Robin: I wouldn't be the right person to present on this, but would be happy to get that presented
… by a proper person
Eric: +1 to that
Robin: I'll look into it
DanD: if the SCONE-PROTOCL proposal gets traction, we should definitely look at how it should be exposed at the Web layer
DanD: I can arrange a presentation of the work after the BoF outcomes
ChrisN: we're also interested in seeing how that proposal evolve
… liaisons with other organizations is also something worth investing in - 3GPP is looking into advanced media delivery (a video-centric use case)
… which intersects with work in the Media WG for streaming apps
… so a liaison with that particular work item in 3GPP sounds particularly relevant
Sudeep: please send more details
Sudeep: Dom, anything related to your report on AI on the web?
Dom: It's higher level, but there are some questions, e.g., role of hybrid models where parts run on the client and some in the network
… Has interesting privacy and security properties, but hybrid usages are very network dependent
… So I can see some intersections, e.g., with the cloud-edge work
… The paper really raises challenges that need solutions, but not sure yet on the intersection here
Sudeep: Happy to bring this up as a topic here
Eric: Another thing: it's cost-prohibitive to run AI workloads in the cloud, so downloading is a trend that's anticipated
… Microsoft put out a white paper, a 1 bit LLM, if they can get to that, more AI workloads may run on handsets and devices
… Today, most investment is in LLM in the cloud, but you'll have to start moving things to the edge and to devices. A lot of that falls in scope for this group
Sudeep: There's a lot of scope there. If you know anyone working on these topics, please let them know, we can schedule them in
… I'll share the links with you
… Max and Song will prepare the CG charter, I'll connect with Robin
Next meeting
Sudeep: I suggest we meet in one month. I'll arrange a time
[adjourned]