The (very) uneven distribution of DNS root servers on the Internet | Hacker News Reader

The (very) uneven distribution of DNS root servers on the Internet

royal.pingdom.com

43 pointsicehawk18 comments

> We think that the root server distribution should take regional Internet population size into consideration, so we hope the people in charge read this and act accordingly as the Internet continues to grow.

This is really the sum of their analysis? They "think"? The "people in charge"? No analysis of actual performance? No discussion of the fact that caching resolvers mean that most queries never hit the root servers? No discussion of the fact that this is just the distribution of sites that the 13 root server IP addresses map to via anycast, and that each site will have many actual physical machines behind that IP address?

How about doing some measurement to see if there is any problem? Some analysis that discusses what effects this might have? Some discussion of who those "people in charge" are, and what it's suggested that the should do?

This kind of drivel really makes me which that Hacker News had a downvote button for articles.

"you can look directly at how many Internet users there are for each root server site in a region. The lower the number, the better." [citation needed]?

I was under the impression that a diverse geographical distribution (to defend against localized natural disasters and targeted attacks) was the primary goal, rather than a concentration of DNS servers in areas of higher internet population.

There apparently is just one root server for all of Russia, and that singlehandedly is responsible for these numbers. (Edit: 4 actually.)

http://www.root-servers.org/

And Canada only has root servers on the eastern side of the county, and none whatsoever in the west.

But does it really matter? Aren't they just used when there is nothing cached?

The author makes it seem like "the powers that be" have some kind of conspiracy, but it's really up to the local ISP's to devote the resources to run a server, there is no one in charge - you just have to do it and then ask to be added to the list.

On that map there is one in the middle of Russia, but there are 3 more in Moscow. I assume that is also the main internet hub for Russia.

Root servers are used for a lot of traffic, not just when nothing is cached. For the AsiaPacifc stats on the L server: http://dns.icann.org/cgi-bin/dsc-grapher.pl?plot=bynode&...

There's far more than one. Root-servers.org shows clusters of root servers, not single physical boxes. Even then there's instances of at least the F, J, K and L roots in Russia (3 of them in Moscow; one in Novosibirsk).

But in any case, the fact that his source does not show individual servers makes the numbers he is giving pointless. It's also meaningless to show geography, when what matters is topology.

Having ten times as many root servers in Asia, for example, would be pointless if they're spread out on networks that have poor connections too each other and roundtrip times for most users don't improve.

It's quite possible, or likely, that there are imbalances that matter, but his numbers don't show it.

I'd be interested to know more about the topology of the UK domestic internet, is there anywhere where one can find such information?

Quote:

    You can see how the world is divided in two parts here:
    Above average (a bad thing in this case): South America, Middle East, Africa, Asia.
    Below average (a good thing in this case): Europe, North America, Oceania.

Just showing such an obvious division, without commenting on it (industrial vs. post-industrial, "developed" vs. ex-colonies, Western vs. everything-else, etc.) seems almost absurd. There's a thing to be said about the way and the ways the US and Europe control the net, it has been said elsewhere.

why is this at all important?

it's not as if the end user ever contacts the root nameservers, their ISPs nameserver does, and it'll cache the rather limited set of results for hours.

It doesn't matter at all, that's why this article is stupid.

Never mind that. Look at how they are distributed in terms of IP (v4) addresses.

198.41.0.4 192.228.79.201 192.33.4.12 128.8.10.90 192.203.230.10 192.5.5.241 192.112.36.4 128.63.2.53 192.36.148.17 192.58.128.30 193.0.14.129 199.7.83.42 202.12.27.33

Back off to the /3 level. All but two of them (128.8.10.90 and 128.63.2.53) are in 192.0.0.0/3.

Guess what happens if someone fat-fingers a sync connection and adds one with a netmask of /3 instead of /30? If it's in the right part of the net, you lose a whole bunch of stuff, but two of the roots keep working.

This makes for some very interesting troubleshooting. I witnessed this once, and it was bizarre: http://rachelbythebay.com/w/2012/01/08/blackhole/

I wonder what else might be tripped up by this kind of aggregation gone wrong? It seems like spreading them out some more could be useful in some really bad situations.

Edit: also, look at the /4 and /5 level and so on. Most of them have a lot in common, actually.

Interesting that Pingdom would choose to complain about a lack of DNS servers in these regions when they themselves do not offer Pingdom probes in any of these regions less represented by DNS roots

If I knew of a similar offering with more global reach I would switch in an instant.

Am I correct in thinking that the root servers only contain the NS addresses for the TLDs? So, in theory, a recursive DNS resolver only has to contact them once per TLD every 6 hours or so?

Yes. You can see the full lookup path with "dig yourdomain.com +trace" if you are interested

pie charts are the worst charts

Devil's advocate says Why?

Because humans are much worse at estimating the relative size of angles than of lengths. See also:

http://en.wikipedia.org/wiki/Pie_chart#Use.2C_effectiveness_...

Of course, the very worst are 3D pie charts, as the angles are distorted.

They aren't real pie.

Maybe not, but they're certainly the worst pies.

Reply on news.ycombinator.com