Alrond's technoblog

Sun 4 Feb 2007

Ergänzungen zum Test der MVC-Frameworks

Posted by Alrond | Tags: CakePHP : CodeIgniter : Django : framework : Der Leistungsfähigkeitstest : php : Pylons : Ruby on Rails : TurboGears : Zend :

Der Test bestimmt die Leistungsfähigkeit der Frameworks unter gleichen Bedingungen für simple Projekte, so wie „hello world!“
Auf die Vorschläge aus dem vorigen Post habe ich den Test etwas ergänzt, so wurden auch Zend Framework (PHP), CakePHP und ein junges interessantes Pylons (Python) getestet.
PHP Frameworks habe ich sowohl ohne als auch mit dem eAccelerator 0.9.5 getestet.
Ich habe auch den Fehler im Test mit RoR 1.2.1 korrigiert.
Auch auf die Bitte habe TurboGears mit verschiedenen template engines getestet.

I habe ApacheBench C5 zweimal und noch zusätzlich C100 gestartet.

ab -c 5 -n 1000 
ab -c 5 -n 1000 
ab -c 100 -n 10000

CodeIgniter und CakePHP lassen sich mit Accelerator unproblematisch starten und testen. Doch Zend funktioniert in Konsole ganz ordentlich, unter FCGI mit Accelerator aber gibt folgendes aus:

Fatal error: Cannot access protected property Zend_Controller_Front::
$_plugins in /srv/zend/library/Zend/Controller/Front.php on line 130

Dank dem Accelerator gelang es die Arbeit der PHP Fremworks in 4.4-6.4 Mal zu beschleunigen. Deswegen ist CodeIgniter als zweites nach Django geworden.

Pylons hat sich als sehr unkompliziert im Verständnis erwiesen, ist auch ziemlich schnell.

Beim Testieren des RoRs habe ich folgenden Fehler zugelassen:
man musste config/environment.rb ändern in RAILS_GEM_VERSION = '1.2.1'
außerdem habe ich entdeckt, daß man im Controller auch "session :off" zuschreiben kann, damit wesentlich die Arbeit beschleunigen. Also, nach der Korrektur zeigte das Framework wirklich viel bessere Ergebnisse, dabei auch das TurboGears überholend.

Fürs TurboGears habe ich außer standardisiertem Kid die Cheetah, Genshi und Jinja getestet. Für das einfache Texttemplate ist Cheetah 1.35 Mal schneller als Kid.


PS: für alle, die mich wegen Voreingenommenheit beschuldigten – für mich waren alle Frameworks gleich. Nur brauchte ich eins für ein Paar ernsten Projekten und das wichtigste dabei ist Stabilität unter sehr hoher Auslastung und selbstverständlich die Geschwindigkeit. Sogar so ein simples Projekt „hello world“ gibt Vorstellung davon und auch von der Projektierungsschwirigkeit mit Framework. So wurde auch ein Wahl zugunsten Django getroffen. Nicht Mal Python kennend, ich habe mich ziemlich schnell mit Django zurechtgefunden, und schon nach einigen Tagen ein Paar kleinen Projekte gebastelt (darunter dieses Blog auch). Es hat sich auch als flinkste gezeigt! Ehrlich gesagt, ich dachte Pylons kann sich schneller erweisen :-) übrigens, auch ein sehr hübsches und geeignetes Framework.

(50 votes, average: 4.5 out of 5)

 Loading ...

 

Top Posts:

• Der Leistungsfähigkeitstest der 6 führenden Frameworks
• Wo kaufe ich ein Notebook mit Linux?! 13 Hersteller antworten...
• Google + Ubuntu = Love ;)

20 Responses to "Ergänzungen zum Test der MVC-Frameworks"

1. Z3R0 Says:
   February 5th, 2007 at 12:23 a.m.

   все правильно!
   работаю с рельсами, а раньше с пыхом трахался
   "как это кривье написали, так и сравнивают. все честно"
   cadmi - 5 баллов!

2. Sokolov Yura Says:
   February 5th, 2007 at 8:59 a.m.

   Вот теперь большое спасибо :-)
   Alrond, прости за рьяную критику прошлого поста здесь и на OpenNet (там я funny_falcon) :-)

   Правда и здесь хотелось бы узнать, на сколько активно было обращение к БД? но ладно, хватит критики, еще раз спасибо :-)

3. Alrond Says:
   February 5th, 2007 at 9:31 a.m.

   Тестировал без обращения к базе - это тема отдельного тестирования...следующего ;)

4. Marston A. Says:
   February 6th, 2007 at 11:12 p.m.

   Even though they are simulated numbers I thought it might be
   interesting to discuss. Besides the fact that from these tests Django
   seems to be whooping serious butt on all fronts, it shows that Rails
   1.2.1 seems to be trailing in performance (quite a bit in some places)
   in comparison to 1.1.6. Anyone seen the same thing in a production
   site?

   I'm curious if there is anything significantly missing in these tests
   and how Django got such great numbers (Or maybe it is just that
   efficient? I have no real Python/Django experience.). Thoughts?

5. Adrian Holovaty Says:
   February 6th, 2007 at 11:13 p.m.

   Django *is* just that efficient. :-)

   But there's a reason for that, and I bring it up here in the hope of
   fostering a good discussion rather than simply "rah-rahing" Django,
   the Python framework that I'm involved in developing.

   We made fundamental decisions in Django that eschewed magic in favor
   of a bit more explicitness. Plus, performance was baked into the
   framework from Day One, rather than us taking a "we'll make it
   efficient later" stance. As a result, Django doesn't have to do as
   many things dynamically as Rails does.

   The tradeoff is that, in Django, you'll have to use an import
   statement if you want to use a certain bit of code, rather than the
   code magically being available to you. The tradeoff is that, in
   Django, you define your database models in Python code rather than
   having the framework introspect the database at runtime. Overall, the
   tradeoff is that, in Django, you're a tiny bit more explicit.

   Of course, this isn't an "I hate my job, this is so tedious" level of
   explicitness -- Django is, after all, intended to save you a ton of
   time. It's more like a pragmatic level of explicitness. And it has
   some nice side benefits, such as the aforementioned performance
   benefit (which is important) and the fact that explicit code is more
   "discoverable" (i.e., it's easier to poke into the framework to see
   what's *actually happening*).

   If you give Django a shot, I'm pretty sure these differences will jump
   out at you. Whether you appreciate the difference is completely a
   personal decision, and no amount of marketing is going to change your
   mind. (Well, for most people.) The difference between Rails and Django
   programmers seems to be the programmers' tolerance for magic.
   Performance is a nice side benefit of less magic.

6. Ben Bangert Says:
   February 7th, 2007 at 8:42 a.m.

   How did you benchmark Pylons? Are you running stand-alone HTTP server? FastCGI? SCGI? mod_python? Each and every one of those will yield different numbers, and Pylons will run in any of those environments.

   I'm rather baffled by the disparity in your benchmarks, that I've never seen in other benchmarks of Django, TurboGears, and Pylons. Can you post more details about the setup?

7. Pento Says:
   February 7th, 2007 at 9 a.m.

   Спасибо за тестирование!
   Если честно не совсем могу понять почему CodeIn c eAccelerator делает в ТРИ раза CakePHP + eAccelerator.
   А вообще, надо серъёзно присмотреться к питону...

8. Alrond Says:
   February 7th, 2007 at 10:56 a.m.

   2 Ben Bangert:
   I've tested all frameworks in an identical conditions: FCGI+Nginx

   For start of pylons I used dispatch.fcgi:
   #!/usr/local//bin/python
   from paste.deploy import loadapp
   wsgi_app = loadapp('config:/srv/pylons/helloworld/production.ini')
   if __name__ == '__main__':
   from flup.server.fcgi import WSGIServer
   WSGIServer(application=wsgi_app, bindAddress=('127.0.0.1', 6500)).run()

   # cat hello.py
   from helloworld.lib.base import *
   class HelloController(BaseController):
   def index(self):
   c.value = "Hello World"
   return render_response('/hello.myt')

   # cat hello.myt
   <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
   <html>
   <head>
   </head>
   <body>
   <% c.value %>
   </body>
   </html>

   For more details see the first test
   http://www.alrond.com/en/2007/jan/25/...

9. tabo Says:
   February 7th, 2007 at 3:58 p.m.

   What happened with the % memory/cpu eaten by every framework? Those numbers were available in the original test.

10. Philip Jenvey Says:
    February 7th, 2007 at 10 p.m.

    As suggested for TurboGears, the templating engine can be a significant part of a small benchmark like this. I'd love to see the new Mako templating engine used in one of these benchmarks (particularly with Pylons, as Mako will be the default Pylons templating engine soon).

    It would also be interesting to see the frameworks' performance without a templating engine involved at all.

    Django:
    return HTTPResponse('Hello world', mimetype='text/plain')

    Pylons:
    return Response('Hello World', mimetype='text/plain')

    TurboGears (I think):
    @turbogears.expose(content_type='text/plain')
    def index():
    return 'Hello World!'
    
    RoR:
    @response.headers['Content-Type'] = 'text/plain'
    render :text => 'Hello World'

11. Ben Bangert Says:
    February 7th, 2007 at 11:02 p.m.

    As Pylons is very modular, the code itself is very fast, I'd say the same speed as Django for all practical purposes. What your benchmark comes down to is how fast it renders the template, and template rendering times definitely differ.

    Your simple test has Cheetah about 1.3x faster than Kid. In more realistic (slightly more complex) benchmarks, Cheetah is generally 10-15x faster than Kid. If you're choosing a system based on such a simple non-real-world benchmark, you have no idea what you'll actually get.

    Consider the benchmarks for template languages posted on Mako here (http://www.makotemplates.org/). Clearly Myghty is twice as slow as Django templates which definitely explains the disparity your benchmark is showing.

    If you want to even things up, remove the template equation. Have your Django function just "return Response('hello world')" and have the Pylons controller "return Response('hello world')". That will actually give you a significantly more accurate benchmark of the core code in one system compared to the other.

    I look forward to seeing those results. :)

12. gasolin Says:
    February 19th, 2007 at 5:13 p.m.

    Thanks for your sharing, I'd like to sugget a cleaner example code
    which is equivalent to django's :-)

    # more turbogears-ic

    from turbogears import controllers, expose

    class Root(controllers.RootController):
    @expose(template="tg.templates.world")
    def index(self):
    return dict(hello_world="Hello World!") # eqal to
    {'hello_world': "Hello World!"}

    the kid template is:

    <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
    <html xmlns="http://www.w3.org/1999/xhtml" xmlns:py="http://purl.org/
    kid/ns#">
    <body>
    <div py:content="hello_world"></div>
    </body>
    </html>

    logging, <?pyhton ?> and "py:extend" statement are not essential for
    turbogears,
    and they may take extra time in this performance evaluation.

    You may interest that in newer version (>1.0b3) TurboGears have better
    default quickstart template, which will lead you less redundant
    code :-)

    With python 2.5 I believe python frameworks could have even better
    performance (tg's py 2.5 support is in svn though, but django & pylons
    are already py 2.5 compatible)

13. unixvps Says:
    August 28th, 2007 at 5:07 p.m.

    hi,

    Thanks for your test! python looks very good. The big plus of php is ability to encode source code using zend safeguard. It's a pity but python doesn't have it.

    Regards,
    unixvps

14. larin Says:
    October 24th, 2007 at 12:21 p.m.

    Даже после таких результатов теста, мне сложно уйти с ПХП, все таки он и привычнее... да и если потр...ся с настройкой сервака, он не будет сильно уступать в скорости Питону.... ИМХО

15. FX Poster Says:
    October 26th, 2007 at 2:58 a.m.

    2 larin
    Будет. ;)

16. Andrey P Says:
    October 26th, 2007 at 8:45 p.m.

    а где Solar, LIMB, Symfony????
    Взять 2 самых медленных FW на php и сравнивать их глупо....
    ЗЫ хотелось бы увидеть тесты

    <?php
    echo'hello world!';
    ?>

17. Alrond Says:
    October 27th, 2007 at 12:05 p.m.

    Про симфонию можно в предыдущей статье посмотреть. Он даже по сравнению с CodeIgniter оказался медленным. Так что не было смысла сравнивать его с ускорителем.
    И еше раз повторюсь: это тестирование MVC-фреймворков, а не мини-кода. Думаешь, что перл или питон на таком миникоде будет медленнее PHP? :)

18. phasma Says:
    December 19th, 2007 at 12:33 a.m.

    Слушай, а можно в следующий раз побольше тестов Pylons? Он же как и TB может разные шаблоны.

19. dvarkin Says:
    March 30th, 2008 at 10:18 p.m.

    очень интересно было бы увидеть результаты с малюткой web.py... без render template

20. Pavel Says:
    April 9th, 2008 at 11:37 a.m.

    Очень интересные результаты, спасибо!

    Когда будет повторый тест ?
    ZF обновился и работает из консоли на ура.

    Протестируй связку PHP + XCache + FCGI + Nginx, будешь приятно удивлен.

Post a comment