What to do about openDNS
Scott Haneda
talklists at newgeo.com
Wed Jan 21 11:08:14 UTC 2009
On Jan 21, 2009, at 1:48 AM, Sten Carlsen wrote:
> Are you really sure this is ALL the fault of opendns?
Mostly, and in my tests, I believe so. However, it was also why I was
asking here, before I go too far out on a limb.
> Seems to me that the addition of www. and other such like stuff is the
> work of various browsers trying to be helpful to their users. If the
> bare domain name does not give an answer, maybe the user was too
> lazy to
> add www., so the browser will try that on his behalf. Search domains
> in
> resolv.conf might also be "helpful".
Absolutely correct. And if you could tell your browser to stop adding
in the www, you reduce this problem by 50%. With that, email servers
do not add www, they just look for a very specific host name. Any
server, client, or whatever, that is using openDNS, will in this case,
query much more than just once or twice.
Remove openDNS from the scenario, and use some other rr, and we drop
right back to nice server behevior.
> You could try to point your own browser to use opendns and see how
> much
> traffic one request for some defunct domain gives and try the same
> with
> dig or host.
That is exactly what I did. It varies a little, but on average,
calling http://example.com through open DNS, assuming example.com was
pointed to my NS, will hit me up 50 times per second. Then it runs
into the browser asking for http://www.example.com and I get it all
over again. At some point, the browser thinks it is nice to try the
non www again, any may even pop over to the www again.
dig does the same, I just did a test, and got 47 hits for one dig
command. I then changed my resolver away from openDNS, and I got 2
queries.
> The solution looks to me to implement some automated script to catch
> the
> domains giving loads of useless traffic and change them to ->
> 127.0.0.1
> or something.
Could be, it is on my mind. More on my mind, assuming this is bad
behavior, is to get openDNS to fix it.
Thanks
--
Scott
More information about the bind-users
mailing list