[time] More fun with Geo DNS

Rob Janssen rob
Sun Nov 19 10:34:14 UTC 2006


Guillaume Filion wrote:
> Hi folks,
>
> In the last few months, I spent a couple nights playing with geographic
> DNS resolution for the pool.ntp.org project.
>
> Some of you may remember that I made a prototype that used BGP tables to
> resolve the closest servers to each client. It didn't work so well,
> because BGP doesn't take latency into consideration.
>
> I decided to try something else this time and go with IP to lat/lon
> conversion and return the closest servers geographically. At first, this
> doesn't seem like a good idea since nothing guarantees that two servers
> in the same city will be at a short network distance from each other. In
> practice, however, it seems that most cities have route exchange centers
> so this works pretty well -- at least in my limited testings.
>
> You can try it out, the addresses are:
> 0.ntp-pool-test.logidac.com
> 1.ntp-pool-test.logidac.com
> 2.ntp-pool-test.logidac.com
> 3.ntp-pool-test.logidac.com
> 4.ntp-pool-test.logidac.com
> and
> ntp-pool-test.logidac.com will return the closest servers in random order.
>   
I think it is an interesting idea.  I tried the above and it returns 
some close servers.
It also returned my own server.  That may be something to try to avoid.

I was wondering how this service gets the data they provide.  So I 
entered some IP addresses
of ADSL connections on my provider's network (xs4all.nl).
It turns out the results are complete rubbish.   They are all cities in 
the country, but not at
all related to the actual location.  It looks like a randomly chosen 
city.  Because this is such
a small country, it does not really matter.  But they might as well have 
returned Amsterdam
or Diemen (the administrative location of the ISP) all the time.

I have visited some websites where advertisements were placed that 
uncannily matched
my real location, so maybe there exist other providers that have better 
data.  However,
my ISP generally does not give out such info.  So it would have to be 
based on survey
forms filled in by other people in the same network neighborhood.
(maybe it is, and maybe the density of known locations is too low so 
that they have
to make the assumption that different /24 networks are still close together)

But again, for this particular application it probably does not really 
matter.

Rob



More information about the pool mailing list