[OpenIPMap] Geolocating remote peerings at IXPs

[Baptiste Jonglez]

On Mon, Apr 13, 2015 at 06:04:21AM -0700, Martin J. Levy wrote:
> Brilliant point; however as the IX is located where the IP address is
> geolocated and the packet must go thru the location where the IX is
> located; I'd say the lines aren't that wrong.
> 
> Isn't the previous hop (the inbound on the networks peering router) still
> correct?

You're completely right.  However, the link towards the following hop
might be incorrect.

Consider the four-hops path below, where the routers in Frankfurt and Lyon
are connected by remote peering through AMS-IX:

Berlin  -->  Frankfurt  --(through AMS-IX fabric)-->  Lyon  -->  Geneva

Since the router in Lyon replies with its AMS-IX addresss, what you will
get with OpenIPMap is either:

  Berlin → Frankfurt → Amsterdam → Geneva

or

  Berlin → Frankfurt → Lyon → Geneva

depending on where you geolocate the third router (either in Amsterdam
according to its AMS-IX address, or in Lyon according to its physical
location).  Both results are incorrect.  The "correct" path would be

  Berlin → Frankfurt → Amsterdam → Lyon → Geneva

but it has more hops (5 hops) than the original traceroute (4 hops)...

Maybe edges could have an optional geolocation as well?  It seems a bit
tricky to compute automatically, though.


> On Mon, Apr 13, 2015 at 5:17 AM, Baptiste Jonglez <bjonglez at illyse.org>
> wrote:
> 
> > Hi,
> >
> > How does OpenIPMap handle remote peerings at IXPs?  Currently, when an
> > address is identified as belonging to an IXP (e.g. AMS-IX), it is
> > geolocated at the location of the IXP (e.g. Amsterdam), and the UI doesn't
> > allow users to change the location.
> >
> > However, due to remote peerings, the actual router might be located
> > somewhere completely different.  For instance, the following addresses:
> > 80.249.211.136 and 2001:7f8:1:0:a500:19:8435:1 belong to the AMS-IX range,
> > but the routers are physically located in Lyon, France (as indicated by
> > latency + external knowledge).
> >
> > Would it make sense to allow editing the location of addresses in IXP
> > ranges?  There might be some issues when two routers doing remote peering
> > appear as consecutive hops in a traceroute with their IXP addresses
> > (unlikely, but maybe possible?).  In that case, the link on the map would
> > indicate a direct connection between the two routers (e.g. Belfast and
> > Lyon), while the actual link goes through the peering fabric (e.g. in
> > Amsterdam).
> >
> > That being said, the problem is more general: the L2 link between any two
> > neighbouring routers might take an arbitrary geographical path, but this
> > is not taken into account by only geolocating routers.  But since it's
> > difficult to have access to this information in general (IXPs are a
> > special case where it's easier), that's probably ok.
> >
> > Thanks,
> > Baptiste
> >
> > _______________________________________________
> > OpenIPMap mailing list
> > OpenIPMap at ripe.net
> > https://www.ripe.net/mailman/listinfo/openipmap
> >
> >
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <https://www.ripe.net/ripe/mail/archives/openipmap/attachments/20150413/50735e31/attachment.sig>