#summary heavy load simulation.

*The results are outdated. The test was conducted on some old version of the application. We are going to repeat the test using new version soon. Stay tuned.*

= Introduction =
It was interesting to check the application throughput on heavy load setting.

= Environment =
|| *Index* || 944429 items including folders and files (average local network), 160Mb on disk ||
|| *Hardware* || AMD Duron 1000Gz, 256Mb, 7200 rpm ||
|| *Software* || Debian GNU/Linux, Java 1.5, Tomcat 5.5 ||

= Tools =
[http://jakarta.apache.org/jmeter/ Apache JMeter]

= Experiments =
10 users, 100 requests for each, 1 sec ramp-up period.

I searched for "potter" in _everything_. It was found, that actual query means nothing for load test results. The single exception is query, which returns no results -- it is evaluated faster. This can be explained by additional routines used by case with non-zero results to represent them and send the _big_ page to the user.

== Resulting (main) graph ==
[http://punksearch.googlecode.com/files/everything-potter_graph.png http://punksearch.googlecode.com/files/everything-potter_graph_thumb.png]

== Distribution graph ==
[http://punksearch.googlecode.com/files/everything-potter_distribution.png http://punksearch.googlecode.com/files/everything-potter_distribution_thumb.png]

= Conclusions =
PUNKSearch shows acceptable performance (even on old hardware). It takes only 100 msec in average to answer a query when 10 users search something simultaneously. Distribution graph reveals good behaviour, since major part of samples are closely grouped.

= Extra =
  * [http://punksearch.googlecode.com/files/PunkSearch_2007-08-13.jmx JMeter test plan used]
  * [http://www.softwareqatest.com/qatweb1.html#LOAD List of load test tools]