GuiGui's Show - Liens
« Is France "keeping local traffic local"? We can't answer that question with RIPE Atlas (as it doesn't measure traffic) but we can measure the path between end-points in France. In our case the end-points are RIPE Atlas probes located in France, and we measure the path with traceroute.

[...]

Currently 755 RIPE Atlas probes are connected in France, which would be over half a million traceroutes. If we only select up to two probes per ASN, we end up with 23k traceroutes, which is much more manageable, and from which we ask the questions:

Do we see IXPs in these paths?
Do we see out-of-country paths? And do they need to be fixed?

[...]

IXP traversal
It is important to note that we cooperated with France-IX and therefore we were able to look at their peering LANs in both Paris and Marseille. The numbers we present here are related to this cooperation. However, we could easily extend this to all of the French IXP LANs if we have the right information.

Lieu                            IPv4 IPv6
France-IX   Paris         27.3%     33.0%
France-IX   Marseille   1.0% 0.5%
Table 1: Percentage of paths that traversed France-IX peering LANs.

[...]

Out-of-country paths

Paths from probes to probes in France that go outside of France are interesting to investigate. There can be valid reasons for such paths. For instance, the collected data contains traceroutes from a probe at CERN that is very close to the Swiss border, where paths are mostly through Geneva in Switzerland. Out-of-country paths could also be caused by sub-optimal routing policies or config errors. These are the paths we'd like to take a closer look at.

For IPv4, we observed out-of-country IP addresses in 9% of the paths, and in 18% for IPv6.

Table 2 below shows a breakdown of the top five countries that are seen in these paths.

IPv4 IPv6
DE 3.7% 7.3%
CH 2.7% 4.5%
GB 2.1% 3.5%
NL 2.0% 6.0%
IT 0.8% 0.7%
Table 2: Specific countries seen in out-of-country paths between selected probes in France.

[...]

IXP and in-country correlation
If IXPs help in keeping local traffic local, it is interesting to look at the correlation between a path going out-of-country and not traversing an IXP. In Figure 3, cells are coloured according to the following four combinations:
 * Path via IXP, not going out-of-country (green)
 * Path not via IXP, not going out-of-country (orange)
 * Path via IXP, going out-of-country (blue)
 * Path not via IXP, going out-of-country (red)

The IXP paths that still go out-of-country (blue) are, in this case, caused by a probe on a LISP testbed. For the other paths that go out-of-country the operators in question could look to see if it would make sense for them to improve this, for instance by peering locally. »

Via le twitter de Stéphane Bortzmeyer (https://twitter.com/bortzmeyer/with_replies)
(Permalink)