Absolutely unscientific comparison of django performance with different caching backends
November 8, 2008 – 02:21
Image via WikipediaSetup: dead simple django application, returning a template stored in a file as a root page. Template has no executable code.
The server is started as FastCGI (10 children) and server through nginx.
Hardware – IBM T42 with 1GB under linux, kde, firefox, thunderbird, emacs, openoffice and bunch of other everyday application open.
Performance is measured with /usr/sbin/ab -n 10000 -c 10 http://localhost:81/
Results are below.
No cache: (dummy:///)
Requests per second: 219.26 [#/sec] (mean)
Time per request: 45.61 [ms] (mean)
Time per request: 4.56 [ms] (mean, across all concurrent requests)
locmem:///
Requests per second: 302.55 [#/sec] (mean)
Time per request: 33.05 [ms] (mean)
Time per request: 3.31 [ms] (mean, across all concurrent requests)
db://cache
Requests per second: 168.21 [#/sec] (mean)
Time per request: 59.45 [ms] (mean)
Time per request: 5.94 [ms] (mean, across all concurrent requests)
file:///tmp
Requests per second: 12.36 [#/sec] (mean)
Time per request: 808.84 [ms] (mean)
Time per request: 80.88 [ms] (mean, across all concurrent requests)
local memcached with 64Mb
Requests per second: 221.29 [#/sec] (mean)
Time per request: 45.19 [ms] (mean)
Time per request: 4.52 [ms] (mean, across all concurrent requests)
Transfer rate: 81.44 [Kbytes/sec] received
I’ll do no analysis, butwhat surprised me most was the (mis)performance of file:/// backend.
