benjojo.co.uk

one honk maybe more

benjojo posted 14 Feb 2024 14:20 +0000

It would appear that Cogent AS174 and NTT AS2914 have de-peered in europe.

All europen traffic between them goes via the USA currently. This means that single homed cogent customers in the EU now have to go via the USA to get traffic to single homed NTT customers. Something that will add at least around 80ms~ of latency.

Here we can see from the NTT looking glass a traceroute to the Cogent DNS Root server it going from frankfurt (where cogent has a anycast node for the root server) going all the way to the US, and if we look at bgp.tools 's super looking glass, we can see all of the communities suggesting that EU networks paths between the two are going via the United States

Screen shot of the NTT looking glass, showing a traceroute going to the USA from frankfurt, and next to is a screenshot of the bgp.tools looking glass with highlights of "AS2914: North American country origins"

hohl@mstdn.at replied 14 Feb 2024 14:19 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/vNV8Xp5W4td6hy4pMW

@benjojo Are there any well-known targets that are single homed behind NTT/AS2914 and located in Europe? Like the C-root servers but from the opposite perspective.

benjojo replied 14 Feb 2024 14:22 +0000
in reply to: https://mstdn.at/users/hohl/statuses/111930269158637413

@hohl I tend to just pick a prefix that is originated by AS2914 (or the T1 target in question) and find a IP that pings.

This can go slightly wrong, but most of the time it's good enough

alarig@hostux.social replied 14 Feb 2024 14:28 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/vNV8Xp5W4td6hy4pMW

@benjojo And it looks like it’s a complete depeering, I see US PTRs on mtr for both directions

notdan@ioc.exchange replied 15 Feb 2024 14:41 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/vNV8Xp5W4td6hy4pMW

@benjojo yikes https://en.wikipedia.org/wiki/Cogent_Communications#Peering_disputes

benjojo replied 15 Feb 2024 16:27 +0000
in reply to: https://ioc.exchange/users/notdan/statuses/111936019798221238

@notdan I can add that to my list of times I've been ref'd on wikipedia!

benjojo replied 14 Feb 2024 14:14 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/vNV8Xp5W4td6hy4pMW

If we look at RIPE Atlas data, we can see that this happened around the 12th of Feb at 18:30 UTC. Complete with a minor blackout between the Cogent and NTT while things were presumably settling down.

25ms to 86ms! Ouch!

A screenshot of RIPE atlas, showing a latency jump from 25ms to 86ms, with a peak at 137ms

benjojo replied 14 Feb 2024 14:51 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/D2Js1tPJ3r7s9XS6Hq

There are a few visible examples of this de-peering causing pain for places that Africa that have a large dependence on Europe for internet connectivity. This traceroute shows a path in Kenya going:

Kenya -> London (cogent) -> (cogent) New York (NTT) -> London (NTT) -> Endpoint.

Looks like this trace has gone from 150ms~ to over 400ms!

A RIPE Atlas traceroute showing Kenya -> London (cogent) -> (cogent) New York (NTT) -> London (NTT) -> Endpoint

benjojo replied 14 Feb 2024 14:52 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/Q2j1V7t7RbBwW82Czv

At the time of writing, there are 948 Single Homed AS174 ASNs, and 105 Single Homed to AS2914, according to the bgp.tools (Paid, sorry!) dependency feature

This means the total impact of this is around ~1098 ASNs or 1.37% of all ASNs.

ASN counts are not everything though... Estimating true traffic impact is a lot harder