Last week, I wrote about graphing JMeter results with Matplotlib. Let's take a closer look at the actual Drupal test plan.
This plan was adapted from Jacob Singh's test and has five different thread groups:
Since some of these samplers need a variety of URLs to test, we have two input files:
urls:
Not surprisingly, this contains a list of URLs for the browsing threads. A simple list can be generated with the following query:
SELECT CONCAT('/node/', nid) FROM node WHERE STATUS = 1 ORDER BY changed LIMIT 1000;
For an active site, an alternative might be to extract this list from your site's access logs; such a list would closely match actual traffic patterns.
wordlist:
The wordlist contains a list of queries for the search thread. Querying the search_index table provides a basic list:
SELECT word FROM search_index ORDER BY RAND() LIMIT 1000;
This list will only contain one-word queries, so again you may want to tailor the wordlist to your site.
To generate multiple samples with different parameters, it's best to run JMeter from a script. In this way configuration options can be parameterized (using the __P()
function) and passed to JMeter with the -J
option. The plotting script in particular expects multiple test runs with a variable number of threads. For this test, the thread_count variable is applied only to anonymous browsing, as it was testing a site that serves primarily anonymous visitors. This balance can be adjusted by editing the thread groups in JMeter.
#!/bin/bash # The host under test. HOST=localhost # A Drupal username. USER=user # USER's password PASS='12345' # A node id to edit. EDIT_ID=42 # Ramp up by factors of sqrt(2). for thread_count in 2 3 4 6 8 11 16 23 32 45 64 91 128 181 256 362 512 do jmeter.sh -n -t DrupalStress.jmx -Jhost=$HOST -Juser=$USER\ -Jpassword=$PASS -Jthreads=$thread_count -Jedit_id=$EDIT_ID done
Each thread group uses a Runtime Controller set to 30 seconds, so this script will take about 10 minutes total. When it's finished, the CSV output will be in /tmp/Drupal6
.