<McCool_> presentation at https://github.com/w3c/wot/tree/master/PRESENTATIONS/2020-10-joint-wot-wn
<scribe> Agenda: https://github.com/w3c/wot/issues/934
mm: goes through the agenda topics on https://github.com/w3c/wot/issues/934
<scribe> scribenick: kaz
kaz for first part
<dom> https://www.w3.org/2020/10/TPAC/wot-update.html#talk
mm: Goal and Use Cases
... work with the existing mechanisms
... shows the diagram of the WoT building blocks
... Current Work Items
... we're on the 2nd Charter
... updates and new topics
... discovery: how to get metadata for IoT purposes
... defined to be global, so not just for the local
networks
... part of this includes directory service
... Thing Description
... based on JSON-LD
... work with external orgs like oneDM
... TD defines abstract data model
<scribe> ... New Work Items
UNKNOWN_SPEAKER: Discovery:
2-phase architecture
... define how TDs are distributed
... Other Activities
... implementation experience
... and Resources
sd: reusing the lightning talk video content
<dom> https://www.w3.org/2020/10/TPAC/wn-update.html
sd: introducing the Chairs
... three Chairs there
... Our Charter
... Mission: explore to leverage network capabilities
... ideas around application hints
... both wireless/wired segments included
... also want to focus on tools
... what kind of metrics to be used
... Our Tasks
... identify opportunities and use cases for network and
application collaborations
... want to see any standardization effort could bridge the
gaps
... for 3G, 4G, 5G networks
... incubation of new work
... and share the latest developments in networking SDOs
... wiki: https://www.w3.org/wiki/Networks
... Progress so far
... (diagram on the group's history)
... 2 work streams
... edge computing and net quality monitoring/prediction
... dig a bit more through the following slides
... Two Workstreams
... network quality monitoring and prediction
... our goal is predict network information from network
infrastructure and operators
... roundtrip time, etc.
... and edge computing
... lot of new use cases there
... offloarding of devices
... a lot of factors to be considered
... regarding when to offloard, etc.
... discovery of edge devices and their capability
... Topics
... network quality monitoring
... network quality prediction
... offload to edge
... trace based network emulation in browser developer
tools
... peer-to-peer eCDN &video cloud service
... Demos
... have a few demos here
<dom> https://www.w3.org/2020/10/TPAC/group-updates.html#webnetworks
sd: Principles for Web &
Networks solutions
... guarantee integrity of the data
... need to keep in mind
... guiding principles
... When something is such a creating medium as the web, the
limits to it are our imagination.
... (by Tim)
... That's it from our side
<dom> Subtopic: Compute Utilities
mm: Problem: Limited Clients,
Sensitive Data
... Trends and Pain points
... privacy and latency are the key problems
... edge computing is counter proposal for that
... would like to combine offload computing
... and also system orchestration
... Use Case Domains: Private
... Home vs Office
... think about private use cases
... robot which has computer vision
... offload to my own PC
... office has similar requirements
... private cloud and offloading
... Use Case Domains: Public
... Retail vs City
... access some service
... IoT devices to be connected
... payment data to be handled
... Smart City services include multi-vendor services
... important to handle privacy constraints
... Edge Compute/Web Hybrids Discussed in W&N
... edge worker
... extend Web Worker to offload work from browser to edge
computer (Intel)
... extend Service Worker to execute computations in the CDN
(CloudFlare)
... distributed browsers
... and mobile worker
... all need "another computer" to offload to
... common requirements for those approaches
... when to offload?
... Edge Worker
... task definition
... moving the task to another environment
... but looks the same
... workload definition onto something
... need to define some utility
... and do the offload to move the work load from one to
another
... the other mechanism of mobile worker has similar
ideas
... Key Issues for Standardization
... discovery of "compute utility"
... common service API for network too
... performance of the network and the utility is
important
... metadata for the performance
... to make decision on offloading is worth doing
... Packaging/Runtime Choices
... container images
... containers allow GPU access
... Relationship to WoT
... discovery of "compute utility" o compute services
... can use WoT Discovery
... semantic/syntactic query available
... should satisfy the need
... but have to have network API
... can describe with WoT Thing Description
... however, still need metadata and metrics
... hard problem is defining the computing performance
... technically we could write it using JSON-based
mechanisms
... Metadata
... metadata about capabilities, performance and
resources
... also about network
... to determine QoS
... relationship between the client the node
... we could query the performance once we could have the
offload mechanism
... need feasibility requirements as well
... challenge for the decide part
... application script on the client side
... want some automated rules
... parameterize the rules
... final issue on identifying whether trustworthy or not
... Metadata
... workload: network QoS, compute QoS
... compute utility
... network (LPP)
... need to have acceleration mechanism like Web GPU
... tricky part is performance
... I have some client and how to handle the offloading?
... compute utility handles memory size, accelerator
technology, performance, etc.
... Offload Decision Rule
... feasible options exist
... users may or may not want offloading
... so would be useful to mention offloading would cost
what
... prbably need some metric for workload
... Summary
... all proposals for edge computing so far need...
... a target to offload to with "good" properties relative to
client
... need to decide whether or not offload would be
beneficial
... decision requires metrics on performance, connectivity,
etc.
... compute utility could be defined that...
... is discoverable
... has standardized network interface
... has standardized workload packaging
... next thing is discussion
sx: what about another short presentation on DNN?
mm: maybe 5mins?
kaz: please let us know about the URL of the resource too
sx: will do
... Overview of executing DNNs on the Mobile Web
... deep neural networks (DNNs)
... DNN execution schemes on the Web
... heavy model vs lightweight model
... Enabling Distributed DNNs for the Mobile Web Over Cloud,
Edge and End Devices
mm: compute utilities could
handle DNNs
... some particular algorithm there
... could be advertised in some way
... you have some particular performance in mind
... there are lots of benefit there
... maybe DNN today but might be another workload possibly in
the future
sx: I see
mm: predefined workload also
would be aligned
... but we're more interested in WoT Thing Description to be
applied general purposes
sx: I see
mm: maybe we need contract kind
of mechanism
... how to best execute the job
... based on the legal requirements as well
... depending on the countries/states
sx: this idea was brought by BUBG
mm: we can figure out the common elements from this proposal too
sudeep: in the current context,
the bigger problem is probably discovery and Thing Description
for edge computing here
... let's discuss that point
mm: yes
... let me go through the WoT Discovery quickly
specifically "4. Architecture"
mm: compute utility can run other
workload
... one of the possible mechanisms for discovery is extending
DNS
... 3 options here
... JSONPath, XPath and SPAQL
... looking at syntactic search mainly and semantic search
optionally
... JSONPath can search for content
... you can have some template for the search
... let's say your workload is some kind of FFT
dan: your zero-trust
assumption
... everything you're looking for could be public
... we need to make clear distinction
... things which the user owns
... vs not owns
... it should address the ability to state what the Thing
is
... what is the incentive?
... use cases with local on my PC, that's fine
... for enterprise, would be similar
... but why we need to accept offloading
dynamically/globally?
... how to accept?
... may be out of the scope but we should have better
description
mm: 3 cases here
... offload to your own thing
... or something managed by the ISP
... or random case
... for example, I have some account managed by OAuth
... give a token to compute
... the third case would be difficult to handle
... so would start with the first one and the second one
dan: yeah
... but even the thing you own should require
authentication
mm: right
... authentication is definitely a requirement
... general problem with edge computing is how to trust the
destination of offloading
... in long term, there would be some solution
... to guarantee privacy as well
dan: tx
mm: different level of trust would be needed here
zk: is there any work on vocabulary for orchestration, etc.?
ack
mm: benchmarks have some
examples
... prototypes for offloading
zk: there is dependency for Thing Description
mm: there is no perfect solution yet
zk: would start with the existing solutions
mm: we can look into what they
do
... monitoring shutdown, etc.
... need PoC
zk: minimal solution would be good
mm: benefit for offloading should be significant
zk: tx
sudeep: quick comment
... when we talk about metadata
... is the metadata for compute utility static or dynamic?
mm: we WoT also have a problem
with geolocation information
... static metadata is easy
... maybe you can query the static metadata first
... and then query the diff for the dynamic data
... geolocation data changes from time to time
... probably you want to update the data periodically
zk: you have support for dynamic range with multiple metrics
sudeep: ok
... the decision on offloading could be handled by the
application
... minimal thing to have
mm: subscribe what to see
sudeep: network api on changes
there
... something similar would be useful
mm: should think about what would
be essential
... how to implement offload rules
... for discovery we have an implementation by Fraunhofer based
on LinkSmart
kaz: probably we should think about concrete use cases for edge computing including the viewpoint of how to deal with offloading
<McCool_> https://github.com/w3c/wot-usecases/pull/31
kaz: I think how to subscribe data and how to manage history of data stream should be also considered there
mm: right
... would like to explain how to deal with use case
proposals
... GitHub repo and raise proposals using Pullrequests
kaz: yeah
... probably we should continue the joint discussion with the
WNIG guys
... probably during the WoT Use Cases calls
mm: right
... that call occurs on Thursday at 7am US Eastern
... would schedule a followup call
... let's continue the discussion by email
sudeep: ok
... we have to talk about metadata for computing too
... if you could provide the detail on the WoT Use Cases call,
we could joint it
mm: ok
... will provide information to you
sudeep: sounds good
[adjourned]
This is scribe.perl Revision of Date Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00) Succeeded: s/Dominuque/Dominique/ Succeeded: s/Gol/Goal/ Succeeded: s/... also problem with packaging// Succeeded: s/zelo/zero/ Present: Kaz_Ashimura Michael_McCool Li_Lin Geun-Hyung_Kim Hiroshi_Ota plh Sudeep_Divakaran Kunihiko_Toumur Dominique_Hazael-Massieux Piers_O'Hanlon Jon_Devlin Dan_Druta Louay_Bassbouss Daniel_Peintner Jack_Dickinson GeunHyung_Kim Martin_Alvarez Song_Xu Cristiano_Aguzzi Zoltan_Kis Tomoaki_Mizushima dezell Kazumasa_Okabe Found ScribeNick: kaz Inferring Scribes: kaz Agenda: https://github.com/w3c/wot/issues/934 WARNING: No date found! Assuming today. (Hint: Specify the W3C IRC log URL, and the date will be determined from that.) Or specify the date like this: <dbooth> Date: 12 Sep 2002 People with action items: WARNING: IRC log location not specified! (You can ignore this warning if you do not want the generated minutes to contain a link to the original IRC log.)[End of scribe.perl diagnostic output]