[dns-operations] "Top 20 DNS server map shows US internet dominance" (pingdom)

[Steve Gibbard]

On Wed, 25 Apr 2007, Duane Wessels wrote:

>
>
> On Wed, 25 Apr 2007, Paul Vixie said:
>
>> http://royal.pingdom.com/?p=125
>
> This is pretty strange for a number of reasons.
>
> Mostly it shows who is in domain squatting/typo/after- market.  Not too
> surprising that the U.S. leads this greedy business.
>
> Second, I'd say there are numerous ways to think of the "size" of
> a nameserver (query rate, hardware requirements, memory usage,
> number of zones served).  And we all know that these squatters can
> and often do serve all their domains with a single wildcard RR.
>
> The last comment about east/west coast is funny too.  Its no surprise
> that popular nameservers are located where there is a lot of Internet
> infrastructure.

Having done something vaguely similar (see the current Internet Protocol 
Journal if you haven't already gotten sick of hearing me talk about 
locations of DNS infrastructure), I understand the impulse to do what they 
did.  However, they're both measuring the wrong things and measuring them 
incorrectly.

Others have already covered how the domains on the biggest name servers in 
terms of number of domains appear to be among the least important 
operationally, so I'll leave that alone.

They claim to be looking at the locations of the DNS servers, but then say 
that they determined the "locations" by looking at the registration 
addresses in whois.  For instance, the two nameservers they identify by 
name, ns1.fabulous.com and ns2.fabulous.com, are listed on their map as 
being in Cary, North Carolina, while when I traceroute to them (from an 
SBC DSL connection in the San Francisco Bay Area), both of those servers 
appear to be in a Savvis datacenter in Santa Clara.

More interesting is the question of what to measure if you wanted to see 
what regions dominated hosting.  Locations of non-parked non-squatted 
websites?  Locations of DNS servers for non-parked non-squatted domains? 
IP addresses assigned to non-DSL non-cable ISPs?  Big traffic destinations 
for port 80 traffic in ISP flow data?

That said, their East Coast/West Coast observation about Internet 
infrastructure in the US appears pretty accurate, even if the data that 
went into it looks all wrong.  I saw the same thing in my look at DNS for 
the root and TLDs.  It's also apparent if you look at where the major US 
exchange points are, so it isn't a surprise to anybody who's spent much 
time working on the Internet.  What is striking is how different the US is 
from Europe in that regard.  Europe has exchange points and root name 
servers spread all over the place, and while it has some that are 
certainly more important than others, it doesn't have anywhere that 
dominates on the level that Ashburn or Palo Alto and San Jose do in the 
US. The US Internet distribution also looks pretty different from the US 
distribution of population, or of other sorts of infrastructure.  Both the 
Internet and the Interstate highway system (to pick one example) have big 
holes in the unpopulated non-Costal West, but the highway system stays 
pretty dense from the East Coast to the Mississippi River, while other 
than in Chicago the Internet mostly sticks to the coasts.

It's also striking how wrong their results look for Europe, or for East 
Asia.  But when measuring something that's only incidentally related to 
what they claim to be studying, I don't suppose that's surprising.

-Steve