Chairs: Tommy Pauly & Marcus Ihlar
Notetaker(s): Caspar Schutijser, Craig Taylor
Speaker: R. Geib
Q&A
Marcus Ihlar - Len Ciavattone has an implementation. Is he up to speed?
[did not get answer]
Ruediger Geib: co-author usually first implements before updating the
draft, but not sure.
Greg Mirsky - "This has been implemented by the open broadband
initiative"
Tommy made a procedural point about "last call"; It wasn't obvious
whether there was an action point: Perhaps he can review.
Speaker: G. Mirsky
Q&A for draft-ietf-ippm-hybrid-two-step
Question from the chairs about readers and reviewers of the draft: About
5x readers in the room and 4 reviewers.
Q&A for draft-ietf-ippm-asymmetrical-pkts
Ruediger: Believes the names is misleading [packets] asymmetrical
packets vs asymmetrical traffic
Greg: leaning toward renaming from asymmetric packets to asymmetric
traffic, and is more than half way through the existing feedback.
Speaker: R. Gandhi
Q&A
Giuseppe Fioccola: comment: add reference to document that I'm
presenting later about the idea to combine active tools and hybrid
methods for hop-by-hop measurements.
Rakesh: we can think about it.
Tommy: as an informative reference?
Giuseppe: yes.
Speaker: B. Tiegen
Q&A
Stuart Cheshire: would be great if you can present this at the IEEE
meeting next week. There's fairly contentious discussion in Wi-Fi
community. IEEE don't want to add CC/reponsiveness data at the protocol
as it's a "layer violation", this helps to inform that conversation.
Tommy Pauly: how are the categories corresponding to particular
application services like WebEx, Zoom, etc.? Were the people
self-indicating what they were running?
Speaker: numbers are hyphotethical if you run a video conferencing tool.
...
Tommy: being able to distinguish between application-platform pairs
(e.g., FaceTime on iOS 18 vs WebEx on Linux) is useful.
Ruediger: the scale is not a moss scale and not reflecting user
experience, correct?
Speaker: correct.
Juan-Carlos Zuniga: correlating to MAC address: there's MAC address
randomization now so be aware of that. Second point: at IEEE next week
we're ineed discussing these topics. Third: they want to know exactly
what you are measuring. Objective.
Abhishek Tiwari: similar median results, more outliers for wifi. users
suffer more often from problems.
Speaker: goal about improving quality is to make the instances of bad
performance more rare
Tommy:
Speaker: a lot of the problems with wifi due to wifi congestion and due
to range
Jason Livingood: turning measurements into actionable recommendations is
what users need.
Gaetan F?: Are you measuring signal strengts and all of those?
Speaker: yes.
(Not sure who cntd.): have you compared router measurements to QoE
measurements provided by user applications/devices?
Speaker: yes, but only on small scale.
Speaker: S. Cheshire
Q&A
Greg Mirsky: interesting point whether this metric allows separation of
network measurement and application measurement. E.g., separate
propagation from queueing.
Speaker: we want to measure things that affect behavior for
applications. so no separation between network and application possible
per se, but we want to minimize impact of application.
Speaker: we want to make it hard to cheat on this benchmark.
Abhishek Tiwari: couple points. Repeatability requirement: some amount
of variance is expected on the Internet.
Speaker: agree. 0 variance is not realistic but a small percentage
variance is realistic. should be low enough so you trust the results of
the measurements.
Abhishek: second point: RPM measurements: surprised to see "throw away 5
% of worst measurements". Outliers are very important.
Speaker: personal view is throwing that away is not the right thing to
do but it does make measurements more repeatable. But it's a workgroup
consensus document.
Warren Kumari: output of tests depends on whether people on same network
(e.g. family member) are using the network heavily.
Speaker: widely held view that performance degrades if people are using
the network. But a video conference can be a 1 Mbit/sec data stream and
it should still work well if other people are using the network. The
Internet is a shared network but we have not learned how to share it
very well.
Warren: agree that's not how the network should be but it is.
Timothy Panton: I can help a bit on how video conferencers think. Two
basic algorithms: one is to estimate bandwidth and push down until you
get to what you can achieve end-to-end. Second: ask for retries until
you get full packet.
Speaker: have heard different perspectives on how video conferencers do
it. need input.
...: ...
Bjorn Teigen: support for not throwing away the worst 5 percent.
Jonathan Lennox: video is complicated and fairly loss-tolerant; audio
less so. Also rexample of glitches that happen that make a bad
experience. E.g. once a minute a Wi-Fi channel scan which turns itself
off for 500 milliseconds.
Speaker: audio probably a better example than video
Speaker: G. Fioccola
Q&A
Greg: propose to look at hybrid two-step. Would like to work with you.
Speaker: thinking of adding reference to your document during your
presentation. Lets' talk offline.
Speaker: G. Fioccola
Q&A
Greg: approproate to work on this. Would like to discuss later: mark
packet ... accurate measurement of delay but also generate operational
information analogous to what's done with ...
Speaker: G. Fioccola
Q&A
Xiao Min: there's a reference to a 6man draft. We now have a new, merged
draft that's replaces the old one.
Speaker: thank you.
Speaker: L. Zhang
Q&A
Greg Minsky: measurements are useful, but measurements can already be
realized [did I get that correctly?].
Speaker: it's hard to make sure everything is measured (e.g., using IP
header information to generate entropy)
Rakesh Gandhi: is it intentional that forwarding nodes have to parse big
part of packets? [did I get that right?]
Xiao Min: current devices do not have the capabilities that you require
4 minutes left, no time left for the lightning talks. Attendees are
encouraged to look at the slides on the datatracker.