Django an Unscientific Benchmark between mod_python / mod_wsgi / aspen and runserver
Information
Tests are executed with Apache's ab2 tool running on the same machine
as the server was. There is no explicit caching done in Django.
CachingMiddleware was not enabled and otherwise default settings where
used.
Test System, workstation:
- core2duo 6700 2 x 2.66GHz
- 2GB Internal Memory
Test Pages
Frontpage View is a custom Django view, rendered as a template of roughly 12kb
in size. It contains several database queries, like a tagcloud, top 3 articles,
top 10 authors and some others. These are not cached, the database is mostly empty
and contains just enough information for the site to work as espected. Database
backend used for the tests is SQLite.
Static template was roughly 15kb in size.
Results
Django 'frontpage' View: ab -n 1000 -c 10 http://localhost/
aspen: 19.71rps django-runserver: 12.07rps apache-mod_python: 32.58rps apache-mod_wsgi-embedded: 36.01rps apache-mod_wsgi-daemon(25 threads): 18.08rps apache-mod_wsgi-daemon(5 processes): 33.63rps apache-mod_wsgi-daemon(15 processes): 35.98rps apache-mod_wsgi-daemon(5 processes 5 threads): 35.56rps
Django 'static' Template: ab -n 1000 -c 10 http://localhost/wj1/
aspen: 111.50rps django-runserver: 66.61rps apache-mod_python: 156.82rps apache-mod_wsgi-embedded: 166.76rps apache-mod_wsgi-daemon(25 threads): 108.97rps apache-mod_wsgi-daemon(5 processes): 191.10rps apache-mod_wsgi-daemon(15 processes): 192.64rps apache-mod_wsgi-daemon(5 processes 5 threads): 189.75rps
TODO
Improvements on this unscientific benchmark
- use postgresql as database backend.
- include full ab2 output.
- also test with some other benchmark tool like httperf.
- use multiple urls, also download static media and other assest to mimic
a user browsing more closely.
- create graphs of the results. (everybody loves graphs)
- use dedicated server
- also include test where run benchmark programs from other system
- include configuration files
- try to meassure multile time points on the server, possible by using aggressive
caching, to identify time spend on database, calculations and django itself.
- benchmark static template on plain webserver (without any django processing)
Information
Tests are executed with Apache's ab2 tool running on the same machine as the server was. There is no explicit caching done in Django. CachingMiddleware was not enabled and otherwise default settings where used.
-
Test System, workstation:
- core2duo 6700 2 x 2.66GHz
- 2GB Internal Memory
Test Pages
Frontpage View is a custom Django view, rendered as a template of roughly 12kb in size. It contains several database queries, like a tagcloud, top 3 articles, top 10 authors and some others. These are not cached, the database is mostly empty and contains just enough information for the site to work as espected. Database backend used for the tests is SQLite.
Static template was roughly 15kb in size.
Results
Django 'frontpage' View: ab -n 1000 -c 10 http://localhost/
| aspen: | 19.71rps |
| django-runserver: | 12.07rps |
| apache-mod_python: | 32.58rps |
| apache-mod_wsgi-embedded: | 36.01rps |
| apache-mod_wsgi-daemon(25 threads): | 18.08rps |
| apache-mod_wsgi-daemon(5 processes): | 33.63rps |
| apache-mod_wsgi-daemon(15 processes): | 35.98rps |
| apache-mod_wsgi-daemon(5 processes 5 threads): | 35.56rps |
Django 'static' Template: ab -n 1000 -c 10 http://localhost/wj1/
| aspen: | 111.50rps |
| django-runserver: | 66.61rps |
| apache-mod_python: | 156.82rps |
| apache-mod_wsgi-embedded: | 166.76rps |
| apache-mod_wsgi-daemon(25 threads): | 108.97rps |
| apache-mod_wsgi-daemon(5 processes): | 191.10rps |
| apache-mod_wsgi-daemon(15 processes): | 192.64rps |
| apache-mod_wsgi-daemon(5 processes 5 threads): | 189.75rps |
TODO
Improvements on this unscientific benchmark
- use postgresql as database backend.
- include full ab2 output.
- also test with some other benchmark tool like httperf.
- use multiple urls, also download static media and other assest to mimic a user browsing more closely.
- create graphs of the results. (everybody loves graphs)
- use dedicated server
- also include test where run benchmark programs from other system
- include configuration files
- try to meassure multile time points on the server, possible by using aggressive caching, to identify time spend on database, calculations and django itself.
- benchmark static template on plain webserver (without any django processing)
