Re: Paper about the web ecosystem

From: S Moonesamy <sm+mu_at_elandsys.com>
Date: Thu, 07 Jul 2016 15:56:10 -0700

Hi Roderick,
At 08:25 07-07-2016, Roderick wrote:
>The 140 GGCs indeed represent the number of Google caches IPs
>included in the responses of the RIPE Atlas DNS queries and that we
>further geolocate with our methodology in Mauritius.
>We are aware of inconsistency in geolocation databases [1]. But as
>of today the 10 data sources we used in this work are the only open
>ones that can be adopted for such purposes.
>To carry out a sanity check, we even performed ping measurements
>with RIPE Atlas probes to select as country of the IP the one from
>which the mean of the RTTs to the said IP is the lowest.
>We however do not have access to any ground truth data against which
>we can compare such results. Thus, the 140 GGC IPs are the results
>of the best effort in geolocation that can be done with open data
>sources as of today.
>We are currently working on the journal version of the paper which
>would detect inconsistencies in geolocation with existing databases
>by comparing with a proprietary geolocation database.

Thanks for the feedback. I encountered a somewhat similar problem
which I was working on IPv4 statistics for the region. I ended up
doing manual checks when I found out that the results I got did not
match local information. It is unfortunately very difficult to get
access to ground truth data for the continent. Your methodology
actually looks good on a first review.

>Moreover, we only study the way IPs allocated by AFRINIC (and not
>all RIRs) are served by Google/CDNs. That is the reason why it may
>appear that the number of caches found in Mauritius is higher than
>the one in Other regions (or countries).

Yes.

>In this case, Google caches in the United Kingdom are shared to serve Africa.
>We will look further into the dataset to shed more light on
>Mauritius in the extended version of this paper.

For at least one network the closer cache seem to be in the United
Kingdom. I unfortunately do not have enough data to show whether it
is an unusual occurrence. I would be interested to read the extended
version of the paper.

Regards,
S. Moonesamy
Received on Thu Jul 07 2016 - 22:57:40 PST