[Pdns-users] New recursor install, 50% timeout on first run
cyruspy at gmail.com
Thu Oct 16 12:33:33 UTC 2014
2014-10-13 13:05 GMT-03:00 Ciro Iriarte <cyruspy at gmail.com>:
> 2014-10-13 10:37 GMT-03:00 bert hubert <bert.hubert at netherlabs.nl>:
>> On Mon, Oct 13, 2014 at 10:10:43AM -0300, Ciro Iriarte wrote:
>>> Hi!, I'm running some quick tests on a new Recursor (3.6.1) and I see
>>> ~ 50% query timeouts. It has a good Internet connection but no clients
>>> yet, can this be caused by empty cache?.
>> Yes, on a cold cache, nameservers are a lot slower.
>> You can improve on this by increasing the number of mthreads (max-mthreads)
>> and the number of file descriptors.
>> The best performance comes from a busy nameserver, as outlined on
>> That your second time results were worse is weird, can you do a test with
>> more than 1000 domains, say, 50000?
>> Can you report "rec_control get-all" before and after a run?
>> Perhaps your PowerDNS server is behind NAT and your other server isn't?
> Hi Bert!, I've run again 1000 domain test with logging disabled (just
> in case). The results look pretty much the same:
> Tried a cold cache with 80k domains, it was a disaster, I couldn't let
> it finish.
> The configuration:
> I'll double check with the networking guys, This server will use a
> secondary public IP, but I see that the queries get out with the base
> private IP, so it's possible that the edge NAT is causing this.
> I'll keep you updated. Thanks!
> Ciro Iriarte
Well, wanted to answer this having the issue solved, but the "ticket
is enqueued" with the networking team. So far I could get statistics
from the NAT machine, the server was hitting the 2048 ports limit.
Running several batches without restarting the server got me 100% hit
So, we're moving the server to the outside (public IP) to continue load test.
Thanks once more!
More information about the Pdns-users