Open Source Working Group Minutes RIPE 77

Thursday, 18 October 2018 
WG Co-Chairs: Ondrej Filip and Martin Winter
Scribe: Gerardo Viviers
Status: Draft

A. Administrative Matters - Working Group Chairs

The presentation is available at:
https://ripe77.ripe.net/presentations/102-RIPE77-Opensource-WG-Agenda.pdf

Ondrej and Martin welcomed everyone to the session and presented the agenda. Martin noted that the working group chair elections would take place and that a call for candidates had been made, but there were no nominations. Martin said that Ondrej and himself were willing to continue as WG chairs and that the next elections would be within one year.

Martin pointed out that the minutes from the previous meeting should have been posted on the RIPE website and made a joke regarding finding them. There were no corrections reported, so Martin assumed they were therefore approved.

Martin asked the audience if the time slot of the Open Source WG taking place in parallel to the IoT WG was a problem. There was a raise of hands and Martin noted that there was a considerable amount of people that were unhappy with the chosen time slot. He then asked if the previous time slot in parallel to the Anti-Abuse WG was a problem for the audience. Only one person raised their hand. The WG chairs will look into the situation and discuss changing the time slot again.

B. IRRd version 4: The Why, How, and the Future

Job Snijders (NTT) and Sasha Romijn (DashCare)

The presentation is available at:
https://ripe77.ripe.net/presentations/104-IRRD-version-4.pdf

Peter (no affiliation) asked on IRC if there was any good Python library to parse IRR data like the route6 object?

Sasha Romijn clarified that there is not a Python library and that is why they wrote a RPSL parser themselves. She pointed out that it could be possible to extract the code that does this and make it available as a separate project.

Peter asked if SQL dumps would be published.

Sasha pointed out that it is a question for NTT to answer as it is operational. She doubted who would use the SQL dumps right now and clarified that importing a SQL dump into a database and then mirroring with NRTM was not a very clean solution in her eyes.

Stavros Konstantaras (AMS-IX) asked if the RPSL parser was a full parser and could parse everything that appeared.

Sasha explained that the parser doesn’t parse everything at the moment. She mentioned that it was something they still had to work on and it would depend on the direction they would be taking in phase 2 of the development of IRRd 4.

Stavros requested Sasha to put the parser from IRRd 4 into a separate project, adding that it is something the community needs at the moment. He came back to Sasha’s request for features to implement and asked for an AS-set resolver.

Sasha clarified that this feature is already available in IRRd and it is done on the fly, resulting in a very fast operation.

C. Three Years of Automating Large Scale Networks Using Salt

Mircea Ulinic, Network Systems Lead at Cloudflare

The presentation is available at:
https://ripe77.ripe.net/presentations/113-RIPE77_Three_years_of_automating_large-scale_networks_using_Salt-Mircea_Ulinic.pdf

Cyril Malevanov (Selectel) asked how Cloudflare does interruptable services and gave an example configuring an interface from access to trunk and vice versa.

Mircea clarified that everything you can do from the CLI can be automated and is very fast.

Cyril asked how the service interruption could be made as short as possible.

Mircea acknowledged not having seen this happen and that anything you could do from the CLI could be automated. He added that there should be APIs that should deal with these things and gave an example with Netconf. He repeated that anything you do with the CLI can be done with automation and you shouldn’t notice any downtime.

An attendee noticed that the Napalm log modules were added. He asked how the observability side of Salt along with the Napalm logs is compared to the streaming telemetry he gets off the packet forwarding engine in a router.

Mircea said that there is an overlap between the Napalm logs and the streaming telemetry. He added that ideally the Napalm logs shouldn’t exist at all, but that the reality is that streaming telemetry is not ready yet. He also mentioned he carried out some tests on Junos, stating that only Junos 17 might be ready and noting that hopefully future Junos releases will have better support. Mircea stated that some Napalm logs would not be replaceable by the streaming telemetry.

D. Fast, Simple User-Space Network Functions with Snabb

Andy Wingo, Igalia

The presentation is available at:
https://ripe77.ripe.net/presentations/55-ripe77-2018-snabb-open-source-slides.pdf

Bengt Gorden (Resilans) asked how Snabb copes with the Intel cards with the multiqueue and if it was fully implemented.

Andy explained that they don’t try to depend too much on all firmware features, because firmware is often buggy. He clarified that they implement VMDq filtering based on VLAN and MAC address, additionally supporting the RSS features of the card.

Bengt asked if the control plane was separate from the data plane in the multiqueue.

Andy clarified that it is currently not separated. He asked for Bengt’s experiences in this matter.

Bengt mentioned that they had written a paper ten years ago about gigabit routing and that they had to patch the kernel drivers for Intel cards so they could reach the network card during a DDOS attack.

Andy noted that their approach was to scale for small packets at high rates. He added that some systems have more of an overhead for control packets than others and that his goal is to reach a throughput that doesn’t have a special control case.

E. Lightning Talks

The presentations can be found here:
https://ripe75.ripe.net/archives/video/167/

E1. OpenBGPD/OpenBSD update

Peter Hessler (OpenBSD)

The presentation is available at:
https://ripe77.ripe.net/presentations/143-openbsd-status.pdf

Tom Hill (Bytemark Limited) asked about the changes in rad (router advertisement daemon) and if it behaved properly with CARP (Common Address Redundancy Protocol) failovers.

Peter confirmed that rad is now aware of CARP and is able to change announcements.

Tom Strickx (Cloudflare) asked if there was a limit on the prefix size for SLAAC.

Peter clarified that limitations had been removed and any prefix size was possible.

Gert Doering (SpaceNet AG) praised the implementation of an origins feature in OpenBGPD 6.4.

E2. Routinator 3000, RPKI Relying Part Software Written in Rust

Alex Band and Martin Hoffmann (NLnet Labs)

The presentation is available at:
https://ripe77.ripe.net/presentations/111-routinator-open-source-wg-ripe77-compressed.pdf

Peter (no affiliation) asked on IRC if they also plan to implement the BGP preview functionality that the RIPE RPKI Validator 3 has.

Martin said that it is in their plans for the future.

Peter added that he only wants an API and not a web GUI.

Martin explained that they will only include the features that the people request.

Martin Winter concluded the session and said he hoped to see everyone at RIPE 78.