[CentOS-mirror] New mirror

19.3.2017, 22.01, Dewangga Bachrul Alam kirjoitti:
> Hello Anssi,
>
> Would you like to share how Cent OS team determine the mirror location
> by an IP address? Let's pick an example, softlayer has an IP test on
> Mexico (speedtest.mex01.softlayer.com), but yes, the traffic is
> through US than directly to Mexico. Possible it's IP Transit, since
> MIE (Mexico Internet Exchange) have same traceroute details.
>
> [...]
>
> The good news is, I can get result from atlas probe, 10 probes on
> mexico are active.
>
> Measurements to 204.45.61.61
> https://atlas.ripe.net/measurements/7930303/#!probes
>
> Measurements to mie.mx
> https://atlas.ripe.net/measurements/7930313/#!probes
>
> Any hints from traceroute above? Hope it helps.
>
> Notes:
> Is there any FAQ from mirror team about choosing mirror location
> determine by an IP Address? Correct me if I'm wrong.

The short answer is: It depends.

Longer answer: When Massachusetts Institute of Technology submitted 
their mirror to the list, it was pretty clear that the mirror was 
actually in Massachusetts. For some other mirrors we may do a quick 
lookup from one of the GeoIP providers out there, such as MaxMind. If 
everything matches, good, otherwise we may dig deeper. There are no 
particular guidelines for us at this stage, because the actions depend 
on the observed discrepancies. Usually asking for a clarification from 
the mirror submitter is enough, and we can usually reach a consensus of 
the mirror's location that way.

I'm unsure how relevant mie.mx is for this discussion, so I'm not 
commenting on that at this stage. However, while researching for this 
discussion, I have found that Mexico's ISP are (still) less 
interconnected than in some other countries. It is very well possible 
that some Mexican ISPs communicate with each other by passing their 
packets twice across the U.S. border. Hence my question about which 
Mexican ISPs centos.mexico.machost.co's host peers with, as that might 
shed some light on the observed oddities.

Thanks for the Atlas probes, I had entirely forgotten about that. I 
think it's noteworthy that all those 10 probes from Mexico to 
204.45.61.61 ended up getting routed to Denver. This reinforces my view 
that the mirror does not seem to have a local network presence in 
Mexico. It looks like it's entirely in U.S., but as Mach Host claims 
otherwise, I'm listening to their input on this matter.

As for a FAQ, no, this does not come up frequently enough to warrant a 
FAQ entry. We generally expect mirror admins to know where their servers 
are, but we check the data they submit to make sure there aren't errors. 
In this age of "everything's in the cloud and nobody sees physical 
servers any more", the mirror admins might not even know where their 
servers are. It happens, but we try to work out the differences.