To twist or not to twist, this is always the question.
Or, rather, I always want to use twisted, but can I produce the numbers
and the tests for customers that show it performs as well as
Or, even better: how does one architect a killer twisted web app server
that performs insanely well with high concurrency and blinding speed?
Hmm, It's been a while since I checked out the Nevow code base, so I
just did an svn update, and I see there there is now a wsgi cgi for
mod_python in the examples directory. This is cool, I shall have to
test... Perhaps I can have my cake and eat it too.
Well, I just installed it on a server of mine and have been running
apache bench against the following:
nevow wsgi mod_python code
a nevow example running on the twisted daemon
the AdytumSolutions web page (custom mod_python/publisher mini-cms)
a tomcat example jsp page.
I ran 5000 requests, 20 concurrent requests for each. Here are some
nevow wsgi - Requests per second: 48.96 [#/sec] (mean)
nevow twisted - Requests per second: 64.85 [#/sec] (mean)
simple mod_python app - Requests per second: 824.22 [#/sec] (mean)
tomcat - Requests per second: 288.01 [#/sec] (mean)
I've just spent about an hour reviewing the Nevow example apps (started
up the examples by cd'ing to the examples directory and
running 'twistd -noy examples.tac'). I must say: Donavan,
Matt, and company have done an amazing job in the last year. Truly
impressive. Lots of really good examples that get the creative juices
flowing and jump start one right into seeing how to solve all kinds of
problems with Nevow. In fact, I am so impressed that, if I can learn
how to get as much performance out of Nevow, I would gladly sacrifice
some raw running/processing speed for ease of development and
maintenance. And extensibility. And joy.
I wonder if I can get my clients to see things this way?
Perhaps it's time to hop back on #twisted.web and start asking
questions I'm sure they are quite tired of answering...