IRTF MAPRG minutes for IETF-111 (online)

Date: July 30 2021 Notes from Dave Plonka

Overview & Status - Mirja & Dave (MAPRG) (5 min)

Apart from typical meeting administrivia, the chairs invited submissions to the upcoming IAB Workshop on Measuring Network Quality.

Measuring Shutdowns and more: Internet Society Pulse - Mat Ford (10 mins)

Mat did an "informercial" about ISOC's Pulse platform and solicited for data contributions and feedback or potential partners.

See slides (contact info on last slide):

Question (Martin Duke): (paraphrasing) By Internet shutdowns do you mean lawful ones by authorities or DoS attacks?

Answer (Mat Ford): It is not DDoS attacks or ransomware sort, it's very much the gov't mandated shutdown(s) we're looking at.

QUIC in IPv4 and getting there via HTTP Alt-Svc and DNS HTTPS - Jan Rüth (10 mins)

Jan shared measurement data about deployment of QUIC on IPv4 and how to get to HTTP3 via HTTP headers and the DNS via HTTPS.

See slides:

Question (Martin Duke): What's the significance of the 18-byte connection IDs in your graph (on slide 4)?

Answer (Jan Rüth): (paraphrasing) I'm not quite sure how many bytes we used before that date, but Dmitry from Lightspeed contacted us suggesting we use longer connection IDs (as his services weren't being picked up by our scanners). I think we also did not identify some Google servers when we used shorter connection IDs.

Comment (Martin Duke): That's interesting b/c the spec says servers should accept 8-byte connection IDs.

Feedback from using QUIC's 0-RTT-BDP extension over SATCOM public access - Nicolas Kuhn (10 mins)

Nicolas briefly presented their work on trying to test the 0-RTT-BDP (Bandwidth Delay Product) extention for QUIC with Satellite-based publicly available Internet services.

See slides:

There are also associated I-Ds linked therein (see slide 13).

Question (Dave Plonka): I recall that the AIX operating system they cached the MTU in the kernel routing table; are there recommendations here how long this BDP state information should be kept?

Answer (Nicolas Kuhn): (paraphrasing) We do not have such a recommendation on the duration in the draft but we recommend that, when wanting to reuse the information, that the implementation to perform a safety check to be sure previously-stored values are valid (or whether network conditions have changed, so they are no longer valid).

Multipath TCP Measurement - Oliver Gasser (10 mins)

Oliver gave a brief presention on their work measuring adoption of Multipath TCP (MPTCP).

See slides:

There is an accompanying academic research paper published in IFIP Networking 2021 that includes a lot more deta.
Also see their site:

Question (Martin Duke): (paraphrasing) The variety of things these middleboxes can be quite interesting. If you're terminating the connection vs. something transparent changes what is generally better to do. Whether it's good or bad depends on what sort of middlebox it is.

Answer (Oliver Gasser): (paraphrasing) I agree. There are different use cases depending on wheter middleboxes terminate the connection.

Comparative Latency Under Load Performance of Broadband CPE - Jason Livingood & Sebnem Ozer (10 mins)

Jason and Sebnem presented briefly on their work studying latency with broadband customer premise equipment (CPE), from January 2020 (onward through the pandemic), involving millions of cable modems, to determine how different modems performed with DOCSIS 3.1 and AQM.

See slides:

Comment (Mirja Kuhlewind): this is a good topic to consider submitting to the [aforementioned] IAB Workshop on Measuring Network Quality.

Question (Jonathan Morton (sp?)): (paraphrasing) What AQM is in operation at the CPE and headend? Is ECN enabled in both?

Answer (Sebnem Ozer): (paraphrasing) For the cable modems the AQM is that defined in DOCSIS 3.1. The CMTS at the head-end has its own AQM, with some values (parameters) being optimized and the same for both modems.
There was no ECN, e.g., no CE-marking, but we are working toward having ECN-marking [enabled] in the future.

Answer (Jason Livingood): ... and also making sure we are not bleaching ECN.

Question (Al Morton): (paraphrasing) How does the test stream (that you used) defining working load affect the results. Can you describe the test stream briefly.
Answer (Sebnem Ozer): (paraphrasing) We current use iperf-3 CP (?) load and then netperf for UDP ping latency measurements. We are working on some
Answer (Jason): We were experimenting a bit; there is not a lot of knowledge about what is the right thing(tm) at scale.