[Chicago] Facebook open sources FriendFeed's real-time Python web framework, Tornado

[Garrett Smith]

I'm maintaining an informal list of benchmark results of a simple "Hello
World" style app for various web servers. I just updated the list to include
Tornado.

These benchmarks are part of a web framework I'm working on in Erlang called
Landshark. The app must print the test "You're that clever shark, aren't
you" in either HTML or plain text. The numbers are from Apache Benchmark
running for 10 seconds under different concurrency levels.

The numbers below are not rigorous by any stretch. In fact, please consider
them, officially, damned lies.

Based on these lies, however, here's my personal takeaway, at least as far
as the Python servers are concerned:

   - CherryPy is very, very fast and has the upside of being threaded, which
   means you don't need to deal with weird async/event based models (see
   Kumar's post to this thread, which is spot on)

   - Fapws, which spells "Fast Asynchronous Python Web Server" is very, very
   fast indeed, as it's mostly written in C -- however it's event based (used
   libev) so beware of blocking operations (e.g. network or file access) and
   slow Python code, which will bring your numbers down to numbers to mortal
   levels

   - Tornado is an impressive contender here -- solid throughput even at
   high levels of concurrent requests; however, it's only slightly faster than
   CherryPy within "reasonable" levels of concurrent requests (10,000 or less)
   and, if you're a speed freak, you should consider Fapws, which blew the
   field away

Obviously there's a lot behind the numbers, such as configuration, system
under test, etc. Like I said, they're lies. If you have questions or, in
particular, can help me make these go faster with smarter configuration,
I'll happily rerun and publish. If you can make WEBrick (a Ruby server) go
faster than my friend's golden retriever, which is trained to serve dynamic
HTML, I'll buy you a beer.

As a final note, the tests themselves will be available with the Landshark
download, once that's posted, so people can re-run the tests for themselves.
Of course, it's not that hard to run ab against a hello world app.

RPS = requests per second
Time = average time in milliseconds to server each request
Benchmark = `ab -t 10 -c <number of concurrent requests>-r http://localhost`

100 Concurrent Requests
-----------------------
=============== ====== ===========
App Server         RPS   Time (ms)
=============== ====== ===========
Fapws             7174          14
Landshark         4479          22
PHP-5             4191          24
Tomcat 6          3554          28
Tornado           2641          38
CherryPy WSGI     2102          48
Jetty 6            937         107
Django WSGI        785         129
WEBrick             43        2350
=============== ====== ===========

1,000 Concurrent Requests
-------------------------
=============== ====== ===========
App Server         RPS   Time (ms)
=============== ====== ===========
Fapws             5359         187
Landshark         4477         224
PHP 5             3062         345
Tomcat 6          3014         345
Tornado           2452         409
CherryPy WSGI     2126         470
Jetty 6           1095         949
Django WSGI        953        1057
WEBrick             50       20607
=============== ====== ===========

10,000 Concurrent Requests
--------------------------
=============== ====== ===========
App Server         RPS   Time (ms)
=============== ====== ===========
Fapws             5213        1920
Landshark         4239        2361
Tomcat 6          2369        4752
Tornado           2265        4439
PHP 5             2239        4282
CherryPy WSGI     1731        5786
Jetty 6            794       13210
Django WSGI        890       12330
WEBrick             84      163425
=============== ====== ===========

20,000 Concurrent Requests
--------------------------
=============== ====== ===========
App Server         RPS   Time (ms)
=============== ====== ===========
Fapws             4788        4178
Landshark         2936        6823
Tornado           2153        9315
PHP 5             1728       11578
Tomcat 6          1362       15102
CherryPy WSGI     1294       15477
Django WSGI        790       27633
Jetty 6            616       32492
WEBrick             63      462464
=============== ====== ===========


On Thu, Sep 10, 2009 at 12:52 PM, Kumar McMillan <kumar.mcmillan at gmail.com>
wrote:
> This may be of interest to fellow Chipynauts:
> http://developers.facebook.com/news.php?blog=1&story=301
>
> "FriendFeed's web server is a relatively simple, non-blocking web
> server written in Python. The FriendFeed application is written using
> a web framework that looks a bit like web.py or Google's webapp, but
> with additional tools and optimizations to take advantage of the
> non-blocking web server and tools."
>
> Docs: http://www.tornadoweb.org/documentation
>
> It uses epoll (from Linux kernel), interesting.
> _______________________________________________
> Chicago mailing list
> Chicago at python.org
> http://mail.python.org/mailman/listinfo/chicago
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/chicago/attachments/20090914/934542c7/attachment.htm>