I attended the 78th RIPE meeting in Reykjavik, Iceland.
The meeting ran May 20th -- 24th, I attended 21st -- 24th.
- Meeting page: https://ripe78.ripe.net/
- Meeting calendar: https://ripe78.ripe.net/programme/meeting-plan/
- Presentation archive: https://ripe78.ripe.net/presentations/presentation-archive/
Ondřej Surý (ISC) & Petr Špaček (CZ.NIC)
In general, this is an effort to clean up DNS behaviour.
In 2019 there was the DNS flag day, after which we'd expect no need for clients to fallback to queries without EDNS.
In 2020, they propose another flag day to modify DNS behaviour so that:
- fragments are eliminated, and in doing so:
- TCP port 53 is opened up more commonly on firewalls
- which relates heavily to: IPv6 fragmentation being way less reliable than IPv4 fragmentation, and the future adoption of DNS-over-TLS.
- additionally defaults EDNS buffer size options to values that won't lead to fragmentation and yet do better than a 512 byte payload (e.g., 1220 bytes)
Brenden Kuerbis and Milton Mueller (Georgia Tech) are looking at the economics of existing IPv4 networks going dual-stack (rather than new networks who can design from scratch). Not the first look at this, but perhaps a stronger economics view rather than network scientists trying to do economics. Nice graphs on growth models.
Observation:
- per-capita GDP explains half of variation in ipv6 capability.
- some fluff on ipv4 costs
- some fluff on ipv4 transfer markets
- Cloud services are the people buying the v4 space
- IPv4 requirements modeling under dual-stack (separate ipv4/ipv6 networks) and conversion (464xlat, 90% IPv6 devices)
Token question from the audience, "is IPv6 proving to be a market failure?", the response was quite sane: the market seems to be working well! It's pricing v4 space accordingly and moving it toward the most valued applications (i.e., content networks).
- https://ripe78.ripe.net/presentations/21-RIPE-Get-Ready-for-a-Mixed-World_-Economic-Factors-Affecting-IPv6-Deployment.pdf
- https://www.internetgovernance.org/research/the-hidden-standards-war-economic-factors-affecting-ipv6-deployment/
There was one dedicated RACI session, but RACI talks are actually sprinkled through the meeting. The talks we awarded RACI grants to are here: https://ripe78.ripe.net/programme/raci/
Neils presented this a couple times: https://ripe78.ripe.net/presentations/62-Taking-The-High-Route.pdf
"Value-based Routing". In short, the specification of as-sets specified in the database for organisations to self-declare as compliant with, say, GDPR or UN Guiding Principles.
Open question for the floor: does this provide a space for folks to make informed decisions around routing?
This is a temporal study of the impact of IXPs on path lengths, reachability, etc.
IPv6-WG was a fairly standard tour:
https://ripe78.ripe.net/presentations/41-2019-05-23-ipv6-fail.pdf
This is Geoff's talk on IPv6 Reliability Measurements. Some takeaways:
- Stateful transition technologies that involve protocol translation show higher levels of instability
- Translation technologies that require orchestration of DNS and network state are also more unstable
- Happy Eyeballs papers over a lot of cracks, including path incongruencies; if your IPv6 path is sufficiently long for Happy Eyeballs to trip over to IPv4, then something is wrong.
- IPv6 fragmentation rate is absurd (>20% TCP, >35% UDP)
Experience with networks such as FOSDEM leans towards IPv6-only + NAT64 being a reasonable approach to wifi networks
- https://ripe78.ripe.net/wp-content/uploads/presentations/117-RIPE78_ERNW_IPv6_Hotspots.pdf
- Elsewhere, Enno also presented on his experience testing platforms in IPv6-only wifi networks: https://ripe78.ripe.net/wp-content/uploads/presentations/42-ERNW_RIPE78_LightningTalk_2019_WiFi_v6only.pdf
- ** personal note, this aligns with our NAT64 + IPv6-only work from 2015 or so, and the current setups for big traveling networks like the RIPE or the IETF networks
A little bit of a discussion around SLAAC's reaction to renumbering events:
- CPE crash or reboot may not have sufficient state to expire an old subnet, which "sticks" on other devices on the network, possibly for a long time
- not a problem in a stable-prefix environment (but that has other privacy concerns)
- this is being discussed also in 6man/v6ops
- https://ripe78.ripe.net/wp-content/uploads/presentations/107-slides-104-v6ops-slaacs-reaction-to-renumbering-events-00.pdf
- suggestion is for hosts to deprioritise prefixes that have ceased to be advertised by a router but have not yet expired (additionally useful would be updates to RFCs to permit shorter lifetimes and therefore more rapid invalidation)
Geoff's routing talk: https://ripe78.ripe.net/wp-content/uploads/presentations/39-2019-05-23-bgp2018.pdf
Mainly:
- IPv4 exhaustion becomes visible in the extent of advertised address space
- IPv6 apparently has fewer "ghosts", the distinction he sees in IPv4 table sizes between Routeviews (North American) and RIS (often EU) which he attributes to "ghost" routes that haven't been mopped up. (I'd posit this may actually be traffic steering internal to the US market, but I don't know.)
- BGP performance largely the same as every, updates propagate most of the way across the network in ~50 seconds
https://ripe78.ripe.net/wp-content/uploads/presentations/113-routing_security_ripe78_snijders.pdf
Job strongly in favour of strict route validation and rejection of routes with invalid origins per the RPKI.
Some big networks such as ATT do this already; some networks reject invalids via peers, some via customers, some via any EBGP session.
Nice web validator: https://www.ripe.net/s/rpki-test
I didn't attend this, but there's a proposal to double the size of the IPv4 IXP pool: https://ripe78.ripe.net/presentations/76-revised-ixp-assignments.pdf
This is the balance between small buffers and large buffers, and how they interplay with streaming services and perhaps most importantly user satisfaction
Katarzyna Wasielewska (Elblag, Poland), Available Bandwidth Estimation Problem Network Calculus in Practice
https://ripe78.ripe.net/wp-content/uploads/presentations/127-Wasielewska_RIPE2019.pdf
https://ripe78.ripe.net/wp-content/uploads/presentations/141-ripe78_preso.pdf
A short look at latency between middle-eastern countries who commonly don't peer between each other, instead routing traffic primarily via the EU
https://ripe78.ripe.net/wp-content/uploads/presentations/142-ripe78_mat_eraofencryption.pdf
This is another look at the spin bit proposal for end-to-end latency measurement with IETF-QUIC and how it works
https://ripe78.ripe.net/wp-content/uploads/presentations/88-MAT-WG-RIPE-78_v3.pdf
A status update on Atlas VM anchors, UI/UX improvements, new hardware probes, future software probes, DNS cookie support, inter-RIR cooperation on RIPEstat, RIS Live
https://ripe78.ripe.net/wp-content/uploads/presentations/133-Danilo_Giordano_MAT_pdf.pdf
A look at domestic internet usage patterns according to five years of flow data: traffic consumption, site popularity, protocol changes (HTTP to HTTPS to SPDY to HTTP/2 to QUIC, etc), CDN expansion and deployment
There were a couple of nice higher-level items that were good primer sessions for a broad audience. Those were:
- Modern Router Architectures, Ignas Bagdonas et al, https://ripe78.ripe.net/presentations/135-modern-router-architectures-190523-final-2.pdf
- Peering Economics 101, Susan Forney (Hurricane Electric) https://ripe78.ripe.net/wp-content/uploads/presentations/93-Peering-Economics-101-RIPE-78.pdf