MAT Working Group Minutes - RIPE 78

23 May 2019, 16:00 - 17:30
WG co-Chairs: Nina Bargisen; Brian Trammell
Scribe: Alun Davies

1. Introduction

Welcome, Scribe, Jabber, Agenda, Minutes

Brian Trammell opened the meeting

2. Five Years at the Edge

Danilo Giordano, Polytechnic University of Turin

Video available at: https://ripe78.ripe.net/archives/video/123

No questions.

3. Dissecting the Speed-of-Internet of the Middle East

Massimo Candela, NTT Communications

Video available at: https://ripe78.ripe.net/archives/video/126

Dumitru Ciobarcianu (Adobe) asked how much of the difference in speed could be caused by lawful intercept in these countries?

Massimo said that they tried to look into this by looking if there are points where the paths go to all the time. So far they could not find any evidence, but it could well be.

Johannes Deger (University Ulm) asked if the researchers differentiated between IPv4 and IPv6 and if there were any differences. Massimo responded that the study is still ongoing. So far they have not detected any significant differences, but they will have to analyse this further.

Nina Bargisen (Netflix) pointed out this links nicely with the Connect WG: it's definitely an issue. She added you cannot just go and peer in the Middle East, because of the circumstances there. Great to have such studies. It might persuade players in the region so that they have better internet.

Ayoub Friaa (Cloud Temple) asked how the researchers can be sure that the latency is not caused by a bad device instead of the actual distance between the measurement points. Massimo said that they did not really look at the geolocation, but mostly security. But they know where the RIPE Atlas probes are. However, they did not check if there are other devices in between.

4. Buffer Sizing Observations at Netflix

Te-Yuan Huang, Netflix

Video available at: https://ripe78.ripe.net/archives/video/128

Ernesto Ovcharenko (LinkedIn) confirmed that TY's conclusions are very much in line with other tests he has seen. Big buffers are not enough. Small buffers result in better user experience.

Brian Trammell asked how the QoE metrics are generated. TY confirmed that they are doing objective scores and that they are correlated with other data.

5. Available Bandwidth Estimation Problem - Network Calculus in Practice

Katarzyna Wasielewska, State University of Applied Sciences in Elblag

Video available at: https://ripe78.ripe.net/archives/video/130

No questions

6. Measurements in the Era of Encryption: Case QUIC Spin Bit

Jari Arkko, Brian Trammel, IETF QUIC WG

Video available at: https://ripe78.ripe.net/archives/video/132

Brian Trammell asked about the implementation status of the spin bit. Jari responded that there about 15 different QUIC implementations, maybe 5 of them have spin bit, some others have promised to have it soon. He said he is optimistic that we have enough to at least do some measurements.

7. RIPE NCC Tools Update

Chris Amin, RIPE NCC

Video available at: https://ripe78.ripe.net/archives/video/133

Jen wondered if it is possible to put RIPE Atlas probes on mobile phones. At the moment we're mostly measuring residential networks (and mostly those of geeks). Kaveh Ranjbar clarified anchors are located in data centres, so those measurements are not going towards residential networks. Chris responded that this is definitely possible in principle, but would require some extra work.

Massimo said that the RIS stream live is amazing. Even though it is still a prototype, it is very useful for research. He urged everyone to ask the RIPE NCC to keep developing all these good things. Randy Bush seconded this. In response to Jen's request he reminded people that there was a RIPE NCC hackathon where they actually built a mobile RIPE Atlas probe.

Daniel Karrenberg asked if the idea is to co-locate software probes with existing hardware probes and see and analyse the differences and the results. Chris said that this would definitely be part of that. Daniel said in this case, it would be great if those people who already host a hardware probe could volunteer to also host a software probe so one can compare results. He added that putting measurement software on phones is only part of the problem. Maintaining the measurement results is the harder problem. This will cost money.

Randy suggested - in response to the scaling problem Daniel mentioned earlier and Jen's suggestion to measure phones - to do IPv6 only and require DHCPv6.