Revision as of 14:47, 15 October 2005

Comparison

Environment

The test environment consisted of two different machines both running Debian 3.1 (Sarge) with the latest official patches applied. The following packages were used:

PowerDNS 

Version 2.9.18-svn (rev 474, 2005-09-03)

PowerDNS OpenDBX Backend 

Version from 2005-10-15

OpenDBX 

Version 0.9.5

MySQL 

Version 4.1.11a, including libmysqlclient14

PostgreSQL 

Version 7.4.7

SQLite3 

Version 3.2.1

One hosted the PowerDNS server while the other was responsible for running the test suite (queryperf is included in the BIND sources):

PowerDNS + database server 

VIA C3 533MHz, 256MB RAM

benchmark client 

Pentium M 1.5GHz, 512MB RAM

Both machines were connected by a 100MBit network and they were the only ones attached to the hub.

Settings

All test were done with default settings for each database - no further optimizations were applied except if stated otherwise.

To get raw database and backend performance, caching in PowerDNS was switched of generally. Otherwise we would get much higher but false results due to packet caching done by PowerDNS. The "slave-cycle-interval" parameter was set to a value high enough so checking for unfresh slaves had no negative influence:

• cache-ttl=0
• negquery-cache-ttl=0
• query-cache-ttl=0
• recursive-cache-ttl=0
• master=yes
• slave-cycle-interval=300

Test description

The test was based on the records in the example.com zone available in the regression-tests directory of PowerDNS. The zone file was converted by zone2sql (part of the PowerDNS distribution) to a set of SQL queries inserted into the database tables created by specific "create table" statements for each DBMS (available along with the OpenDBX backend sources).

The speed of lookups was tested by running queryperf with a query set of 10000 A records (host-0.example.com to host-9999.example.com) three times in a row.

The AXFR (zone transfer) test was done by running "host -l example.com <serverip> 1>/dev/null" 100 times in a loop while timing start and end of the test. Possible variation of the results due to the operating system should be minimal.

Results

Back to Overview