Broken domain statistics...

Dalvenjah FoxFire dalvenjah at dal.net
Wed Feb 11 21:24:24 UTC 1998


Hi all,

I'd finally had it up to here with people coming from misconfigured
domains trying to connect to my servers and filling up the logs with
'Host name mismatch..' errors, so I decided to put together a bunch
of scripts to try and see exactly how widespread the problem of bogus
DNS info is.

What I found was kind of surprising. Here're the raw stats for my test
(done on the entire com zone):

Total domains checked:						1401150
Domains with NO good nameservers (all responded non-auth):	236008
Domains with NO good nameservers (some timed out):		107482
Domains with at least one bad non-auth (but most/all answering):99211

Bringing that into percentages, about 17% of all domains in COM have
NO good nameservers listed. If one adds in the nameservers that timed
out, that number goes up to 25%, and adding in domains with at
least one bad nameserver brings the number up to 32% of the domains
in .com that have bad nameserver info registered.

Please note, it appears that it's not entirely accurate to view the
nameservers that timed out as necessarily 'bad' in my test - several
known good nameservers timed out during the runs, and I only had a
retry of 1 (so the nameservers got 1 chance to give the correct data
within 4 seconds).

It's very interesting, however, that the number of domains that
had all listed nameservers respond, all of which responded
non-authoritatively (i.e. 'I don't know about this domain') is so high.

Here is my testing methodology:

I read in the entire com zone, and when I found a line containing
   dom	IN NS server
I would spawn off a 'dig ANY dom @server +retry=1', and parse the
output to see if it contained 'aa', the authoritative flag. If it
did, it was a good domain. If it didn't, it was a bad domain. Timeouts
and non-authoritative responses were counted separately.

I then had three variables for each domain - goodResponses, badResponses,
and timeouts. Domains where badResponses and timeouts were both 0 were
considered 'good'. Domains where goodResponses and timeouts were both 0
were considered to have 'no good nameservers (but all responded)'.
Else, domains where goodResponses was 0 were considered bad (noting
that some queries timed out). Beyond that, if there was 1 or more
badResponses, it was listed in the 'at least one bad non-auth NS'
list.

The processes that did this would fork out about 80 processes per host
to run the digs; I was on a reasonably fast connection, so bandwidth
shouldn't have become a problem as far as increasing the timeouts I got.
I split the com zone file into 200,000 line sections and ran one section
per host. I then stopped the stats collection server after every few
runs to gather statistics.

The queries occurred over a period of 12 hours between 3pm and 3am
Pacific Time, Tuesday 2/9 - Wednesday 2/10.

I've put up the code, results, and the logs of non-auth queries and
timed-out queries at ftp1.dal.net:/pub/misc/domain-test/. The files
haven't (and won't) propogate to the mirror sites. (Note, this machine
will be switching IPs sometime this week, so there may be a period of
an hour or two when the machine will be unreachable.)

If I've got some sort of flaw in my logic, please let me know; I'm
willing to correct it and run the test again. But it looks right.

I haven't tested the net/edu/org domains, but I suspect that since
folks using those are slightly more clued than the folks using .com,
the numbers of bogus domains will be lower.

And if it is, it means that 17% of the folks on the internet are
paying for domains that don't work. Either that, or something else
is broken.

I'm posting here because I feel it is an operational issue; that,
plus I feel there're more folks here who can and will hammer at
InterNIC to start doing something to enforce their policies that
require real, authoritative nameservers.

One last request - if you plan to use this data somewhere, *please*
listen to any responses that may show up here explaining how it might
be wrong, and *please* go through my methodology and find out for
yourself if it looks right. I don't want to be responsible for any
false/overinflated claims out there }:P . And please provide context,
too, especially where the 'nameserver timed out' statistics are
concerned.

Anyhow, there it is.

-dalvenjah

-- 
 Dalvenjah FoxFire (aka Sven Nielsen) Stupid people shouldn't breed.
 Founder, the DALnet IRC Network
 e-mail: dalvenjah at dal.net             WWW: http://www.dal.net/~dalvenjah/
 whois: SN90                           Try DALnet! http://www.dal.net/



More information about the NANOG mailing list