Plone performance sprint
Factory tool and types tool revisited
After doing some performance improvements to content creation last week, we are now investigating this further at the Plone core at the Plone Performance Sprint in Copenhagen.
We have been looking for better ways of benchmarking at the sprint, and are currently using JMeter. For these tests we recorded a script where the user logs in, goes to add form, creates the folder, view the new folder and then publish it. JMeter then replayed this script 100 times and generated statistics. This gave us a quick and simple method to see how Plone is performing.
In this test we emulated a customized site by adding 20 content types (copies of regular content types). The user has Manager role, and the machine this is tested on is a MacBook Pro.
First we tested Plone 2.5, as the Factoryhack optimization product we developed earlier could be installed in here directly.
There is a noticeable improvement when viewing the add form and when creating a folder. Because the folder forms use the content listing extensively, the improvements are more noticeable when creating folders compared to other content types. Those wanting to see results for all content types will have to wait until we have a more complete suite of JMeter test plans.
For Plone 3 we had to make a slight adjustment to the types tool patch, as it checks an interface instead of an attribute. Other settings were the same.
Wow, works in Plone 3 as well. It seems like the add form and content creation is slightly slower in Plone 3 compared to Plone 2.5, and the sprinters are currently working on it. Viewing content is faster.
This exercise was mainly about JMeter test plans. Now that we can confirm that the factory and types tool improvements work, those will either be released as a product in the collective (naming scheme still not decided), or included in Plone itself if possible.