[Pdns-users] This can't be right - DNS Load Test Results

bert hubert bert.hubert at netherlabs.nl
Fri Nov 30 23:31:22 UTC 2007


On Fri, Nov 30, 2007 at 11:06:06AM -0800, Max Clark wrote:
> I recently ran a series of dns load tests against various dns servers
> and have been greatly surprised by the results
> (http://www.clarksys.com/2007/11/20/dns_load_testing_results/) - in
> this test PDNS was slightly faster than TinyDNS and half as fast as
> BIND. Because the query set was highly repetitive during the test I
> would have expected PDNS to cache the results and serve subsequent
> queries from this cache. That does not seem to have been the case

Interesting - many people are currently benchmarking PowerDNS. Did we
appear in a magazine or so? It certainly is a trend.

Whatever the reason, we've found an issue where the query cache (not the
packet cache) will not cache backend queries with multiple answers - which
is actually very common.

Additionally, the distributor-thread mechanism scales badly over (modern)
CPUs, which is something we intend to fix very shortly - a matter of days.

I would be very pleased if you could rerun your benchmarks again when the
time is right.

Having said all this - actual powerdns performance problems are very rare in
the real world, and the million domain+ installations keep chugging along
nicely.

But it is nice to have more headroom, and we're working on it.

> daemon=yes
> disable-axfr=yes
> distributor-threads=6
> launch=gmysql
> gmysql-host=10.10.10.10
> gmysql-dbname=dns
> gmysql-user=user
> gmysql-password=pass

It might be interesting to add:

query-cache-ttl=180
negquery-cache-ttl=180
cache-ttl=180
webserver=0.0.0.0

And then check the built-in webserver to see the various cache hitrates on
http://powerdnsserver:8081/

Good luck!


-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services


More information about the Pdns-users mailing list