RIPE 63

Measurements, Analysis and Tools Working Group
RIPE 63, Vienna
3 November 2011, 16:00-17:30

Status: Final

Co-chairs: Richard Barnes, Christian Kaufmann, Ian Meikle
Scribe: Ann Barcomb, RIPE NCC
Chat: Mirjam Kuehne, RIPE NCC

A. Intro

Christian Kaufmann, WG co-chair, opened the session.

Christian's presentation is available at:
http://ripe63.ripe.net/presentations/174-MAT-Administrative.pdf

B. RIPE TTM data in real time - Antonio M. Moreias, NIC.br

Antonio's presentation is available at:
http://ripe63.ripe.net/presentations/170-ttm-real-time-nic-br.pdf

Todd Underwood, Google, asked if it would be possible to release the original data. The question was withdrawn.

C. Measurement Lab - Tiziana Refice, Google

Tiziana's presentation is available at:
http://ripe63.ripe.net/presentations/175M_Lab_presentation_RIPE63_03_11_2011.pdf

Jan Zorz, go6, asked if the host needed to obtain the hardware, or if this was sponsored.

Tiziana stated that it was ideal if the host provided all equipment, but depending on the specific situation, M-Lab is willing to compromise in order to expand the network.

Jan asked what the motivation would be for the host.

Tiziana replied that there are a number of possible motives, such as a desire to get quality data, altruistic support of the Internet, transparency, and a better understanding of problems.

There were questions from two remote audience members about the possibility of integrating with or sponsoring RIPE Atlas.

Tiziana has already scheduled a meeting with Robert Kisteleki to discuss these points.

Ragnar Anfinsen, Altibox AS, asked what speeds can be accurately measured.

Tiziana replied that the platform hadn't been tested with the scenario Ragnar described, and therefore she could not offer details, but the system had been tested on university networks in the US that also had high bandwidth.

Antonio M. Moreias, NIC.br, expressed concern that with open-source clients there is the potential for regulators to apply pressure, and that people could be tempted to fake results.

Tiziana acknowledged the potential for this to happen.She said that it is not a problem now, but could become a challenge when the project grows.Solutions are being considered, but there is a desire to keep it open.

A remote member of the audience asked what technology was used for browser version, and what was the motivation behind the choice.

Tiziana said that they are using primarily Java applets and JavaScript.The tools are built by researchers, who make their own choices about methodology and language. The only requirement is that it must be open-source.

Daniel Karrenberg (RIPE NCC) added to the earlier discussion of Atlas by stating that Atlas is not focusing on broadband measurement because there are others operating in that space.There is some synergy between the projects in other areas, however.He stressed that Atlas should be kept independent from specific big sponsors.

Zulliger Martin Vadovice, Interactiva SL ES, asked about the bandwidth requirements for hosting.

Tiziana said all requirements could be found on the website.

Martin asked if there was a preference for particular types of hosts.

Tiziana referred to documents on the website, and elaborated that there is a preference for highly connected hosts.A node in every ISP is a goal.Because of the desire to expand, all hosts are accepted at the present time.

D. Hurricane Electric World IPv6 Day experience - Martin Levy, Hurricane Electric

Martin's presentation is available at:
http://ripe63.ripe.net/presentations/178-Hurricane_Electric_-_RIPE63_MAT_-_Vienna_-_Nov_2011.pdf

Robert Kisteleki (RIPE NCC) asked if Martin felt that existing path MTU problems will go away, or if this is something Atlas should measure.

Martin replied that it should be measured because it is easy to measure.He does not think that the problem will go away, as it has not done so in the last 10 years since the RFC was created.He suggested continuing the conversation later.

Jen Linkova, Google, expressed the opinion that there will be improvement because of ICMP handling in IPv6.

Martin agreed.

Patrick Gilmore, Akamai, questioned the lack of a Y-axis in the graphs.He wanted to know if there was a sustained and real increase in uptake due to IPv6 day.

Martin replied that the figures were in the order of 10s of gigs of data.They have chosen not to make more detailed graphs public because of the finite number of companies producing the traffic and the possibility of transit or customer information being inferred from the information.V6 traffic is still represents a very small percentage of the total traffic.

Patrick stated that they had also seen an actual increase across the board.

Bendict Stockebrand, Freelance Trainer and Consultant, opined that filtering with IPv6 will hurt so badly that people will have to fix it.

Martin replied that this was a valid point, but noted that it is imperative that anyone starting with v6 should be able to get it working properly from day 0 in order to avoid having a negative experience lead to abandonment of v6.

E. PCAP Analysis with HADOOP - Wolfgang Nagele, RIPE NCC

Wolfgang's presentation is available at:
http://ripe63.ripe.net/presentations/122-RIPE63_WolfgangNagele_Large-scale_PCAP_Data_Analysis_using_Apache_Hadoop.pdf

Tiziana Refice, Google, asked what the biggest challenges were.

Wolfgang replied that a library was missing, but it had been trivial to write it.The biggest problem was that Hadoop is a young and developing technology, and it is necessary to have a good understanding of how it works in order to use it in a stable environment.

F. Update in RIPE NCC Activities - Vesna Manojlovic, RIPE NCC

Vesna's presentation is available at:
http://ripe63.ripe.net/presentations/180-ripe63-mat-wg-sd_update-v5.pdf

Questions were postponed until after the following talk.

G. RIPE Atlas User Defined Measurements, Functionality and Implementation - Robert Kisteleki, RIPE NCC

Robert's presentation is available at:
http://ripe63.ripe.net/presentations/140-ripe63-mat-robert-atlas.pdf

Richard Barnes proposed making that the default option should be to make probes public.

Robert noted the suggestion.

Thomas Schmidt, DFN, asked about scalability and control of resources.

Robert replied that the current plan is to do this with bandwidth limitations.There are also plans to allow people to opt out of particular types of measurements.He stated that he was open to other suggestions.

Martin Levy, Hurricane Electric, referred to an earlier slide on maps of route name servers and asked for the data to be sorted by value.He asked if it will use a UDN mechanism that allows someone to add, for instance, a ccTLD anycast named server or enterprise anycast named server.

Robert answered that this would be the power of Atlas, and that it is what they are aiming at.

Martin asked if this will be public.

Robert replied that it would be public to everyone within two weeks.

Sebastian Wiesinger, Noris Network, asked how probes are selected for global measurements.

Robert said that it is done randomly, but they will try not to overload particular probes.

Sebastian expressed a desire to be able to select probes from a number of different regions.

Robert said he sees this as an advanced level feature, but in line with current lines of thinking.

Wolfgang Nagele, RIPE NCC, referred to the map of anycast deployment. He said that a prototype had been shared with route server operators, and that it was widely agreed that this provides the most comprehensive view of anycast effectiveness seen so far.

An attendee commented that the plans for user-defined tests carried the potential for DDoS attacks and wondered how many requests would be run at the same time.

Robert answered that organizations which host more probes will have more “credits” to run tests on other probes, because the system is based on reciprocity.There will still be the option for hosts to restrict traffic in order to avoid overloading their networks.

Daniel Karrenberg (RIPE NCC) further elaborated that scheduling user-defined measurements is a central process, and there is a rate-limiting possibility per destination.This design decision was based on two considerations: first, the possibility of DDoS attacks, and secondly, to prevent redundant measurements on particular probes of interest on networks which may already be under strain.

Bengt Gorden, Resilans AB, reminded the speaker that he had earlier requested the option of tagging downtime in Atlas in order to provide information he has received from other sources.

Robert responded that it could be useful to tag probe status with information deducted from the environment.A situation where the probe is running, power was present, and there was local Internet is different from a scenario where the probe reboots.This information is quite valuable.

H. Measuring Prefix Reachability - David Freedman, Claranet

Christian Kaufmann, WG co-chair, extended the scheduled agenda to include a brief speech by David.There were no slides.

David asked the audience how the reachability of a prefix could be measured so that it is possible to build filters properly.He proposed, as an experiment, deliberately announcing an incorrectly registered prefix and observing how far it propagates, in order to measure the lifespan and experience of such prefixes.David stated his intention of formalizing this idea as a proposal.

Wolfgang Nagele (RIPE NCC) mentioned a presentation he had done earlier about announcing anchor prefixes from RIS with one valid ROA and one invalid.

David clarified that he was not talking about ROAs.

Richard Barnes stated his interest in the idea and advised that David bring this to the MAT WG mailing list.