[{"author": "Randy Bush", "text": "

they are doing it to make folk very glad we did not go to philly

", "time": "2022-07-26T14:00:08Z"}, {"author": "Marwan Fayed", "text": "

Masks are mandatory!

", "time": "2022-07-26T14:01:33Z"}, {"author": "Randy Bush", "text": "

remotely, we see a title slide

", "time": "2022-07-26T14:02:37Z"}, {"author": "Juliusz Chroboczek", "text": "

Am I the only one who hears noise behind the speaker's voice?

", "time": "2022-07-26T14:04:27Z"}, {"author": "David Oran", "text": "

I'm getting distortion too

", "time": "2022-07-26T14:04:38Z"}, {"author": "Michael T\u00fcxen", "text": "

Same for me

", "time": "2022-07-26T14:05:23Z"}, {"author": "Juliusz Chroboczek", "text": "

Yes!

", "time": "2022-07-26T14:05:32Z"}, {"author": "David Oran", "text": "

yes

", "time": "2022-07-26T14:05:32Z"}, {"author": "Jonathan Hoyland", "text": "

Much better

", "time": "2022-07-26T14:05:32Z"}, {"author": "Yi Huang", "text": "

better

", "time": "2022-07-26T14:05:33Z"}, {"author": "Michael T\u00fcxen", "text": "

Better

", "time": "2022-07-26T14:05:35Z"}, {"author": "Oliver Borchert", "text": "

yes bettwe

", "time": "2022-07-26T14:05:42Z"}, {"author": "Juliusz Chroboczek", "text": "

TJ?

", "time": "2022-07-26T14:07:27Z"}, {"author": "Colin Perkins", "text": "

Tijay Chung from Virginia Tech, the co-chair of the workshop

", "time": "2022-07-26T14:07:54Z"}, {"author": "Jonathan Hoyland", "text": "

Taejoong \"Tijay\" Chung

", "time": "2022-07-26T14:08:24Z"}, {"author": "Juliusz Chroboczek", "text": "

TY

", "time": "2022-07-26T14:08:31Z"}, {"author": "Randy Bush", "text": "

@lucas: yes

", "time": "2022-07-26T14:10:09Z"}, {"author": "Lars Eggert", "text": "

we see a black screen in the room

", "time": "2022-07-26T14:12:09Z"}, {"author": "Petr \u0160pa\u010dek", "text": "

We can see black screen with a cursor.

", "time": "2022-07-26T14:12:15Z"}, {"author": "Alessandro Amirante", "text": "

now they're available

", "time": "2022-07-26T14:16:54Z"}, {"author": "Juliusz Chroboczek", "text": "

(I was hoping it was a reference to 8\u00bd.)

", "time": "2022-07-26T14:19:20Z"}, {"author": "Jonathan Hoyland", "text": "

I thought the IETF was layer 8 :P

", "time": "2022-07-26T14:19:27Z"}, {"author": "Jonathan Hoyland", "text": "

But no 5

", "time": "2022-07-26T14:20:27Z"}, {"author": "David Oran", "text": "

In fact there's no [latform 5 in the map he's showing!

", "time": "2022-07-26T14:21:06Z"}, {"author": "Juliusz Chroboczek", "text": "

They've got 8 tracks numbered from 0 to 8. Unlike TCP/IP, which has 5 layers numbered 1 to 7 ;-)

", "time": "2022-07-26T14:21:52Z"}, {"author": "Chris Box", "text": "

Currently Cloudflare Radar is telling me 0% HTTP, 100% HTTPS. More precision needed!

", "time": "2022-07-26T14:31:29Z"}, {"author": "Jonathan Hoyland", "text": "

If you hover-over the tooltip text gives you 6 decimal places.

", "time": "2022-07-26T14:34:54Z"}, {"author": "Jonathan Hoyland", "text": "

I'm seeing HTTP: 0.485541% vs HTTPS: 99.514459%

", "time": "2022-07-26T14:35:43Z"}, {"author": "George Michaelson", "text": "

That was me, (george) as Dobby

", "time": "2022-07-26T15:00:00Z"}, {"author": "George Michaelson", "text": "

delay and jitter are fundamental properties of the network and link layer(s) dependent on the properties of the physical layer capabilities under noise and contention

", "time": "2022-07-26T15:00:57Z"}, {"author": "Gorry Fairhurst", "text": "

The audio of the seaker seems very low...

", "time": "2022-07-26T15:01:00Z"}, {"author": "St\u00e9phane Dodeller", "text": "

same here

", "time": "2022-07-26T15:01:12Z"}, {"author": "George Michaelson", "text": "

but that aside, a very entertaining keynote, most enjoyable. the metrics you have in CF for 4and a bit and above , are fascinating

", "time": "2022-07-26T15:01:33Z"}, {"author": "Alessandro Amirante", "text": "

I boosted a bit the audio level

", "time": "2022-07-26T15:02:31Z"}, {"author": "Alessandro Amirante", "text": "

should be better now

", "time": "2022-07-26T15:02:37Z"}, {"author": "Randy Bush", "text": "

it is

", "time": "2022-07-26T15:02:46Z"}, {"author": "Al Morton", "text": "

it is thanks

", "time": "2022-07-26T15:02:49Z"}, {"author": "St\u00e9phane Dodeller", "text": "

thank you!

", "time": "2022-07-26T15:02:59Z"}, {"author": "Marwan Fayed", "text": "

Hi Folks. I'm told the wifi in the room is flaky. If slides drop again, I'll take slide control and advance on 'next slide pls' command.

", "time": "2022-07-26T15:06:15Z"}, {"author": "David Oran", "text": "

q: what gives you confidence that the results for 2 flows competing extend to N flows competing?

", "time": "2022-07-26T15:08:29Z"}, {"author": "George Michaelson", "text": "

David Oran said:

\n
\n

q: what gives you confidence that the results for 2 flows competing extend to N flows competing?

\n
\n

A good Q, because the experiental sense is that one BBR pushes out many Cubic, where many Cubic seem to share equitably and two BBR can share equitably but I am unsure 3 can, or 3 BBR and 1 Cubic can share equitably

", "time": "2022-07-26T15:12:16Z"}, {"author": "Lucas Pardue", "text": "

@george Michaelson I agree the lower layers play a huge role. I suspected the IAB workshop might focus on those aspects and ignore the potential for upper layers to ruin end-user experience in spite of all the good work being done below. The link to our IAB position paper is https://www.iab.org/wp-content/IAB-uploads/2021/09/Lower-layer-performance-is-not-indicative-of-upper-layer-success-20210906-00-1.pdf in case that helps

", "time": "2022-07-26T15:13:12Z"}, {"author": "George Michaelson", "text": "

Lucas Pardue said:

\n
\n

@george Michaelson I agree the lower layers play a huge role. I suspected the IAB workshop might focus on those aspects and ignore the potential for upper layers to ruin end-user experience in spite of all the good work being done below. The link to our IAB position paper is https://www.iab.org/wp-content/IAB-uploads/2021/09/Lower-layer-performance-is-not-indicative-of-upper-layer-success-20210906-00-1.pdf in case that helps

\n
\n

I also thought that from a 2/3 layer view, QUIC has more qualities of session than transport sometimes. So, the 4/7 thing might really be 4.5/7 as the intrusion of session stability under variant transports, and then I think JSON is presentation layer.. so I kind of think you're in 4.5/6/7 land.. its beginning to get crowded up there.

", "time": "2022-07-26T15:15:11Z"}, {"author": "Spencer Dawkins", "text": "

I would also love for other fairness metrics to gain traction, especially if they work. :grinning:

", "time": "2022-07-26T15:18:21Z"}, {"author": "Lucas Pardue", "text": "

interesting, in my experience you can't handle JSON until the application protocol has terminated. So my reality is more that lots happens above layer 7 and therefore beyond the normal concerns of the IP comms stack

", "time": "2022-07-26T15:19:34Z"}, {"author": "Juliusz Chroboczek", "text": "

George Michaelson: it depends on your application, I guess. If you're doing multiple short HTTP requests, then your session layer lives above HTTP. If your whole session consists of a single QUIC connection, then parts of QUIC do indeed serve as a session layer.

", "time": "2022-07-26T15:25:58Z"}, {"author": "Spencer Dawkins", "text": "

George Michaelson said:

\n
\n

I also thought that from a 2/3 layer view, QUIC has more qualities of session than transport sometimes.

\n
\n

I agree, especially with the multipath QUIC extension, but even with RFC 9000 connection migration, we're beyond transport, as I understand transport.

", "time": "2022-07-26T15:26:27Z"}, {"author": "George Michaelson", "text": "

As a ML skeptic I am actually really delighted when people get traction on ML applications like this. The one presented yesterday in IRTF on ML in the PISA model for switching doing per-packet forwarding logic was really interesting. As was this one. Thanks for presenting. I suspect a LOT of work went into the problem classification and ML training on the data from 2 years of outages. You had access to an awesome list of information.

", "time": "2022-07-26T15:33:16Z"}, {"author": "Christine Pukropski", "text": "

George Michaelson said:

\n
\n

As a ML skeptic I am actually really delighted when people get traction on ML applications like this. The one presented yesterday in IRTF on ML in the PISA model for switching doing per-packet forwarding logic was really interesting. As was this one. Thanks for presenting. I suspect a LOT of work went into the problem classification and ML training on the data from 2 years of outages. You had access to an awesome list of information.

\n
\n

same! cool to see. the one yesterday too was interesting.

", "time": "2022-07-26T15:34:50Z"}, {"author": "Gorry Fairhurst", "text": "

I wonder what type of satellite link was used, that saw propagation loss error, rather than ACM modulation/coding changes at the physical layer?

", "time": "2022-07-26T15:37:34Z"}, {"author": "Juliusz Chroboczek", "text": "

A simulated one, slide 15

", "time": "2022-07-26T15:38:43Z"}, {"author": "Jan Evang", "text": "

Thanks for your comments. I think some projects have tried to use ML for tasks where it is not suitable (it is not always predictable) in this project it has proved to be a very valuable tool. Any mis-classification would have little impact.

", "time": "2022-07-26T15:38:57Z"}, {"author": "Gorry Fairhurst", "text": "

Yeah, I gues this is a GEO satellite service - but which sort of satellite system has 1% loss from propagation?

", "time": "2022-07-26T15:40:22Z"}, {"author": "Michael T\u00fcxen", "text": "

Which loss rates are typical for GEO sattellites (I don't know much about satellite communication)?

", "time": "2022-07-26T15:41:25Z"}, {"author": "George Michaelson", "text": "

I am tempted to ask if they could repeat for LEO on a shorter RTT, similar asymmetry. Only because SpaceX suggests there's going to be a LOT of uptake in LEO, and its a fertile space for aggressive content delivery competition for the bandwidth where I think GEO sat has maxed out its likely use for interactive. Great for Unidirectional.

", "time": "2022-07-26T15:42:31Z"}, {"author": "George Michaelson", "text": "

Nice paper. well presented, clear, easy to understand slides, good logical ordering of the story. Thanks for making it easier for me to understand and follow.

", "time": "2022-07-26T15:43:20Z"}, {"author": "Gorry Fairhurst", "text": "

LEO would be interesting - but there's no common understanding of the path charactacteristics as far as I know - the LEO systems are evolving. I think GEO will continue ... modern big satellites have can dramatically change the cost/capacity tradeoffs ...

", "time": "2022-07-26T15:45:26Z"}, {"author": "Juliusz Chroboczek", "text": "

Would anyone happen to know whether the satellite operators are deploying AQMs?

", "time": "2022-07-26T15:45:43Z"}, {"author": "Gorry Fairhurst", "text": "

There has been research projects to do that. There are complications in buffering for multiple access systems and AQMs which make this an interesting problem space!

", "time": "2022-07-26T15:47:05Z"}, {"author": "George Michaelson", "text": "

Gorry Fairhurst said:

\n
\n

LEO would be interesting - but there's no common understanding of the path charactacteristics as far as I know - the LEO systems are evolving. I think GEO will continue ... modern big satellites have can dramatically change the cost/capacity tradeoffs ...

\n
\n

I think there's a lot in this: GEO is well understood, so is a better basis for reproducible research right now. Also stable: they're big investments up there which arent changing much. LEO is moving rapidly into new spaces like inter-sat comms, and are changing all the time because of rapid deployment of smarter devices. Right now I'm told they're just bent pipes. But not forever.

", "time": "2022-07-26T15:50:28Z"}, {"author": "George Michaelson", "text": "

Big thank you to the ANRW PC for selecting nice talks. I was only going to graze into this one but I wound up staying all the time, because they were interesting. So you did a good PC filter on content, you picked nice ones. Well done.

", "time": "2022-07-26T15:55:12Z"}, {"author": "Gorry Fairhurst", "text": "

On QUIC ACKS, you might like to look at draft-fairhurst-quic-ack-scaling, and see whether this sort of change would be similar to what you saw in picoquic. This ref also provides some background: DOI: 10.1109/ASMS/SPSC48805.2020.9268894; get in touch if you have questions.:grinning:

", "time": "2022-07-26T15:55:17Z"}, {"author": "George Michaelson", "text": "

(rather unfortunate diagram having 312 \"longer\" than 320 ms in the lines. The figures told a better story than the visuals on the FEC cost)

", "time": "2022-07-26T15:57:12Z"}, {"author": "Lucas Pardue", "text": "

thanks everybody, nice session

", "time": "2022-07-26T15:58:25Z"}]