RIPE Labs post: Does The Internet Route Around Damage?
Dear colleagues, We analysed last week's AMS-IX outage from the perspective of RIPE Atlas, to shed some light on the question if the Internet routes around damage. Our analysis is here: https://labs.ripe.net/author/emileaben/does-the-internet-route-around-damage... Kind regards, Emile Aben RIPE NCC
We analysed last week's AMS-IX outage from the perspective of RIPE Atlas, to shed some light on the question if the Internet routes around damage.
Our analysis is here: https://labs.ripe.net/author/emileaben/does-the-internet-route-around-damage...
Great analysis, thanks! A few questions/observations/suggestions for further analysis by aspiring routing researchers... On IXP-less backup paths: "The main alternative in our data are paths without IXP. This could be a direct lateral peering that is not IXP-mediated, or using transit instead of the IXP." Would it be possible to distinguish between those cases? Maybe by looking at any additional ASes that show up in the new paths, and trying to detect provider/customer relationships using some known approach. Intuitively the second case seems more "normal", because if you had a direct peering (PNI), then why would you have preferred AMS-IX in the first place? But of course it's possible that those non-IXP-mediated peerings are over slower links or otherwise inferior. On the belated NL-IX rerouting: This indeed looks like an intentional change of routing preferences; looking at the steep/vertical increases, probably done by a single or a few large providers. Now I'm of course very curious who did this (and was able to shift about 15% of all Atlas traceroutes around :-). Cheers, -- Simon.
Hi Simon, Thanks for your suggestions for further research. More inline :) On 2023-11-29 17:26, Simon Leinen wrote:
We analysed last week's AMS-IX outage from the perspective of RIPE Atlas, to shed some light on the question if the Internet routes around damage. Our analysis is here: https://labs.ripe.net/author/emileaben/does-the-internet-route-around-damage... Great analysis, thanks! A few questions/observations/suggestions for further analysis by aspiring routing researchers...
On IXP-less backup paths:
"The main alternative in our data are paths without IXP. This could be a direct lateral peering that is not IXP-mediated, or using transit instead of the IXP."
Would it be possible to distinguish between those cases? Maybe by looking at any additional ASes that show up in the new paths, and trying to detect provider/customer relationships using some known approach.
Yes, I'm curious about that too. For ease of analysis I don't do AS lookups (just IXP LAN lookups) for this type of use-case so analyses take significantly less time to run. Adding AS-lookups would open up the type of analysis you mention.
Intuitively the second case seems more "normal", because if you had a direct peering (PNI), then why would you have preferred AMS-IX in the first place? But of course it's possible that those non-IXP-mediated peerings are over slower links or otherwise inferior.
Thanks for that intuition, that helps trying to understand this esp. for individuals like me who don't configure BGP interdomain routing every day.
On the belated NL-IX rerouting:
This indeed looks like an intentional change of routing preferences; looking at the steep/vertical increases, probably done by a single or a few large providers. Now I'm of course very curious who did this (and was able to shift about 15% of all Atlas traceroutes around :-).
Or it is an issue of the bias of the probes and destinations that I used for this analysis. This is an issue I'm currently looking into. Happy to share more details about the analysis. kind regards, Emile Aben RIPE NCC
Thanks for the great work as usual. Would it be possible to extend it by looking at measurements other than traceroute that could get us even better insights into what proportion of packets did not get there and how quick the rerouting happened. I assume that adding other measurement types would significantly increase the number of data points and the time granularity. To amplify: i am interested in the dark red areas here. Daniel --- Sent from a handheld device.
On 29. Nov 2023, at 13:54, Emile Aben <emile.aben@ripe.net> wrote:
Dear colleagues,
We analysed last week's AMS-IX outage from the perspective of RIPE Atlas, to shed some light on the question if the Internet routes around damage.
Our analysis is here: https://labs.ripe.net/author/emileaben/does-the-internet-route-around-damage...
Kind regards, Emile Aben RIPE NCC
--
To unsubscribe from this mailing list, get a password reminder, or change your subscription options, please visit: https://lists.ripe.net/mailman/listinfo/routing-wg
Hi Daniel, On 2023-11-29 18:27, Daniel Karrenberg wrote:
Thanks for the great work as usual.
Would it be possible to extend it by looking at measurements other than traceroute that could get us even better insights into what proportion of packets did not get there and how quick the rerouting happened. I assume that adding other measurement types would significantly increase the number of data points and the time granularity.
Thanks for the suggestion. For the error-rate part the analysis doesn't need traceroute, so anything that measures end-to-end between the selected source/destination pairs can indeed be used as an estimator for the error-rate. Most obvious would be pings that run more frequently than the traceroutes on RIPE Atlas typically. kind regards, Emile Aben RIPE NCC
To amplify: i am interested in the dark red areas here.
Daniel
--- Sent from a handheld device.
On 29. Nov 2023, at 13:54, Emile Aben <emile.aben@ripe.net> wrote:
Dear colleagues,
We analysed last week's AMS-IX outage from the perspective of RIPE Atlas, to shed some light on the question if the Internet routes around damage.
Our analysis is here: https://labs.ripe.net/author/emileaben/does-the-internet-route-around-damage...
Kind regards, Emile Aben RIPE NCC
--
To unsubscribe from this mailing list, get a password reminder, or change your subscription options, please visit: https://lists.ripe.net/mailman/listinfo/routing-wg
participants (3)
-
Daniel Karrenberg -
Emile Aben -
Simon Leinen