Test Suite Performance

Events happening in the community are now at Drupal community events on www.drupal.org.
jeffsheltren's picture

I was trying to gauge what the best setup for a PIFR node is, and I noticed that this page (http://qa.drupal.org/node/62) seemed way out of date, so I ran some tests of my own. Here I was trying to figure out a couple of things:
- Which concurrency setting is best?
- How does performance improve with number of cores

For testing, I used two separate PIFR nodes running in KVM virtual machines with the following specs:

CPUs are Opteron 6172, 2.1 Ghz
2 GB RAM
Local disk (10K SAS)
Running Debian Squeeze 64-bit, installed PIFR using DamZ's puppet setup

My results were as follows:

With 8 CPUs:
Concurrency = 9, enable test took 25:23

With 4 CPUs:
Concurrency = 3, enable test took 49:06 (retest took: 46:29)
Concurrency = 4, enable test took 37:20
Concurrency = 5, enable test took 38:20

With 2 CPUs:
Concurrency = 2, enable test took 1:08:01
Concurrency = 3, enable test took 1:10:13

Note that http://qa.drupal.org/performance-tuning-tips-for-D7 claims concurrency should be set to [number of CPU cores] + 1 (e.g. set to 5 for a 4-core system). From my tests with 2- and 4-core systems, it looks like it is actually better to set concurrency to be equal to the number of cores. Also, improvement isn't linear with number of cores -- rfay had mentioned this was the case, but I wanted to be sure. So doubling the cores doesn't quite double the speed (though it does get "close").

Comments

Thanks for this!

rfay's picture

Super helpful, real numbers. Thanks, Jeff.

Drupal.org Testing Infrastructure

Group notifications

This group offers an RSS feed. Or subscribe to these personalized, sitewide feeds:

Hot content this week