[Pdns-users] Benchmarking PDNS backends

[Norbert Sendetzky]

Hi all

To get some facts about speed, I've done some benchmarking on the bind
and ldap backend. The bind backend should be ok as reference for
comparing different machines.

For testing I used queryperf from the bind package on my computer and
all software - queryperf, pdns and openldap - was running on the same
machine. The queryperf readme recommends to do it on two machines,
but I don't have such a setup available.

I decided to do only a backend comparison and therefore disabled all
caches in pdns. Furthermore all queries are of type ARecord and
generated by a script included in the package below. This is a little
bit artificial, but gives much more compareable results.

My setup:
- Celeron 600MHz, 256MB RAM
- Debian 3.0r2
- PDNS 2.9.12 (ldapbackend from cvs)
- pdns.conf without caches (see queryperf/input/pdns.conf)
- pdns is started from command line (pdns-server --config-dir=/tmp)
- OpenLDAP 2.0.23 (pres,sub,eq index on associatedDomain attribute)
- queryperf package is attached
- 1000 test records (file in queryperf/input/)
- 10000 queries x 10 runs

My results:
- bind backend 4087 - 4671 qps
- ldap backend 225 qps

It would be very exciting, if as many users as possible could run the
same test with different backends for comparison. Everything you need
(record file, pdns.conf) to do the same test is attached.

You are encouraged to tune your backend as much as possible, except
caching all queries. This should be a benchmark for testing cold
queries (requests for records that are not cached somewhere) to get
the raw performance of the backend.


Norbert
-------------- next part --------------
A non-text attachment was scrubbed...
Name: queryperf.tar.gz
Type: application/x-tgz
Size: 27820 bytes
Desc: not available
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20031210/951de08b/attachment.bin>