Skip to main content

MAT Working Group Minutes - RIPE 88

Date: Tuesday, 21 May 2024, 16:00 - 17:30 (UTC+2)
Chairs: Massimo Candela, Nina Bargisen, Stephen Strowes
Scribe: Stephen Suess
Status: Draft

View the recordings

View the stenography transcript

View the chat logs

Session Introduction

Stephen Strowes welcomed the attendees, introduced the co-chairs Massimo and Nina, and outlined the agenda, which includes talks on inter-domain routing, RIPE Atlas usage, and Starlink constellations. He encouraged participants to engage via Meetecho and to rate presentations.

Replication of 20 Years of Inter-Domain Routing Policy Modeling

Savvas Kastanaki

The presentation is available at:
https://ripe88.ripe.net/presentations/44-RIPE88-20-Years-of-Inferring-Inter-domain-Routing-Policies.pdf

Savvas focused on modelling inter-domain routing, examining whether the cornerstone model from 2003 still applies today using updated data and methodologies. The study found changes in the consistency of routing policies, particularly with peer routes, and emphasised the importance of understanding selective announcements and their persistence. 

Randy Bush asked Savvas to comment on a 2013 paper, a survey of inter‑domain routing policies by Gill Shapira and Goldberg which measured that the four models held for only about 65% of the relationships.

Savvas replied that there were differences in methodology, he used looking glass servers and not surveys, highlighting the automation in his work.

Antonio Prado asked whether they had considered using RFC 9234 and thanked Savvas for his analysis

Savvas said that he did not recall that RFC.

Pavlos Sermpezis asked whether people changed how they defined their policies or whether it has also changed our view of the Internet. He said in the past we had a view of the Internet and we build a model on that, but today perhaps we want to see other parts of the Internet that we didn’t see in the past.

Savvas said that there are many confounding factors that play a role in these changes. The biggest reason is probably the transition from a hierarchical model used back in 20o3 with the large ISPs holding a lot of power. With the emergence of CDNs, the Internet topology has flattened. A lot of ASes have more aggressive peering policies because peering is less expensive or they have better performance.

Massimo thanked Savvas and invited the next speaker.

Whois Right? An Analysis of Whois and RDAP Consistency

Simon Fernandez

The presentation is available at:
https://ripe88.ripe.net/presentations/54-slides.pdf

The next talk (WHOIS Right An Analysis of WHOIS and RDAP Consistency) by Simon Fernandez, a postdoc researcher , focused on the consistency of Whois and RDAP (Registration Data Access Protocol) records. He analysed the accuracy and reliability of these protocols in providing registration information for domains, highlighting issues with inconsistencies between them. The study found significant discrepancies, especially in name servers, and determined that RDAP was more reliable than Whois in the majority of cases. However, inconsistencies still remain, and there is no absolute way to know which data is accurate without additional verification. The research emphasises the need for careful use of these protocols for domain registration information.

Stephen Strowes said that Simon had identified a lot of mismatches. He asked if Simon had approached operators to correct these mismatches.

Simon responded that some operators were contacted in special cases. Some of those contacted fixed the set up. It was hard to know if it was a registry problem or a configuration problem.

Moritz Mueller (Remote) asked if Simon looked at name servers at the parent or child level.

Simon responded that they focused on parent level for reliability, the child level nodes are inconsistent and hard to get.

Analysis of Usage Patterns in RIPE Atlas Measurements

Pavlos Sermpezis

The presentation is available at:

https://ripe88.ripe.net/presentations/45-RIPE88_MAT_AI4NetMon_Sermpezis_final.pdf

Pavlos Sermpezis presentation discussed the RIPE Atlas tool developed as part of the AI4NetMon project, which aims to identify and quantify biases in Internet measurements. The tool analyses patterns in RIPE Atlas measurements, focusing on probe usage and ASN representation, rather than measurement results. It helps users detect biases and understand measurement patterns. The tool, still in development, is open-source and seeks user feedback for further improvements. It's designed for network operators and researchers to gain deeper insights into measurements and optimise their approaches.

Robert Kisteleki, RIPE NCC, thanked Pavlos for his work and said that he saw it as complementing what is built into the system, and was useful from a different perspective. He asked about the sustainability of the tool and whether it was a long-term project.

Pavlos responded that the project was funded thanks to the RIPE NCC Community Projects Fund and they have built an API, an automated database and a web app tool. They planned to maintain it for a few more years but they are open to collaboration if someone wanted to help make it more sustainable.

Massimo asked if the tool can be exported like open source or used by another organisation in case Pavlos no longer wanted to maintain it.

Pavlos confirmed that it was open source.

Lai Yi Ohlsen, Measurement Lab, said that it would be interesting to have the same analysis on Measurement Lab data or other open datasets. She asked if Pavlos had considered extending the tool to identify when the biases lead to false positives or negatives.

Pavlos replied that a clarification was needed about bias, that it is a characteristic of measurements. You might choose to run an auto-biased measurement for a specific purpose but it does not mean that you are wrong. However, in some cases you might be unaware that your measurement is biased and you might want to fix it. The tool is helpful to fix this.

Eric Lanfer

The presentation is available at:

https://ripe88.ripe.net/presentations/43-RIPE88_Presentation_elanfer_v2.pdf

Eric Lanfer discussed his team's research on Starlink's network performance, focusing on two datasets: WetLinks and Starlink on the road. WetLinks involved stationary measurements in two European cities, capturing data over six months and incorporating weather data. They found that rain impacts download speeds. The mobile dataset, collected via a van in Germany, revealed that movement and obstructions affect throughput. Key findings include better stationary performance, differences between dish generations, and power consumption challenges in mobile setups. Future improvements include direct DC power usage and possibly adding solar panels.

Stephen Strowes asked about reducing power consumption of Starlink dishes.

Eric replied that they converted power from DC to AC and back to DC. It’s possible to have some efficiency there by attaching the Starlink dish directly to car consumption on the car network. In summers they are considering putting a solar panel on the roof of a car to harvest energy when the car is standing still.

Daniel Karrenberg said he uses a 12 volt direct power supply for a Starlink dish which doesn’t use the router and that it’s much better. He offered to talk to Eric offline.

Eric thanked Daniel.

Ben Cartwright-Cox mentioned DC only boards to power the POE, but it requires a special kind of POE because it’s a lot of power over POE.

Eric thanked Ben for his comment.

RIPE NCC Update

Robert Kisteleki

The presentation is available at:

https://ripe88.ripe.net/presentations/10-kisteleki-MATWG-RIPE88.pdf

Robert Kisteleki gave an update on RIPE NCC tools work, discussing efficiency gains by moving historical data to Amazon S3 to reduce costs, testing a pilot cloud controller, the new UI and customised dashboards and more. He mentioned the new firmware for easier development and that software probes would be curtailed to prevent misuse. He also encouraged community involvement, particularly commercial users to support RIPE NCC’s measurement activities financially.

Randy Bush, IIJ, asked about the measurements and modelling used to assess cost reductions in moving from self-hosting to AWS.

Robert said that the RIPE NCC would share more details at the next RIPE Meeting, explaining what worked and what didn’t. He couldn’t promise this but it could be a good topic.

Pavel Odintsov asked about the costs and what they were looking for.

Robert referred Pavel to the article highlighted in the presentation which mentioned numbers.

Moritz Mueller, SIDN Labs, said that he recently used RIPE Atlas measurements in Google BigQuery for the first time which was surprisingly inexpensive and also very useful. He asked how frequently the data was being used.

Robert replied that it is in a beta stage and they are looking at value versus cost to determine the future.

Daniel Karrenberg commented that it would be good for the group to take note of the bigger discussions in the RIPE community and RIPE NCC membership about the measurement activities. He said there were more and more people questioning whether the RIPE NCC should be involved in these activities. He added that if they wanted to hold on to these useful tools, they should take note of these discussions and try to influence them or talk to someone in their organisations. Secondly, he asked those using tools like RIPE Atlas in a commercial sense to sponsor it. He said the other discussion on funding becomes untenable if there are people making money off things that are supported by the RIPE NCC membership funds. He warned that change is coming and that they needed to face reality.

Randy Bush said that he was not interested in micromanaging the RIPE NCC or RIPE Atlas, and that he would move to Amsterdam if he was. His question was really how many of us face similar issues of hosting our own services or moving to the cloud. His intuition was that moving to the cloud might cost the same except it would pay Amazon’s salaries. He said that he thought the RIPE NCC had probably done a reasonable analysis and he was interested in the analysis and modelling and that he was not questioning the decision or discussing the RIPE NCC budget.

Robert said that they had indeed looked at the numbers and seen what they could do with what was on-premise. The decision was made based on the general direction of trying to save costs without doing harm to the services offered.

Randy reiterated that he was interested in how that was modelled and measured, not the budget itself.

Finally Massimo Candela made closing remarks talking about mailing list and encouraging people to provide feedback. He thanked his co-chairs, the stenographer and the audience and invited them to the next session in Prague.