[Update 10/6: Slight updates to the Pylons test after some input from Ben. Added a couple new charts for you visual folks.]

As I briefly mentioned in Round 1, this whole thing came about as an experiment to satisfy my own curiosity. Unfortunately, I wasn’t expecting these posts to draw the amount of attention they have been getting, and several people informed me of a few “issues” with the first round. Since my initial approach to this topic was somewhat casual, I didn’t really take the time to perform each test in a “proper scientific fashion.” Although this was clearly stated in the introduction to round one, it unfortunately resulted in performance estimations that were somewhat less than accurate.

After input from various people much smarter than myself, I quickly went to work tweaking my test environment and building “proper” test apps. In the midst of this, a conversation about PHP accelerators prompted me to put PHP under the spotlight, which brought about Round 2 as an interim round. This gave me a chance to demonstrate the necessity of PHP acceleration, and only continued to solidify my opinion of PHP as an inferior web development language (remember, I just said my opinion).

Which brings us to Round 3. A lot of work has gone into “doing it right” this time, so I am fairly confident that these results are a much more accurate representation of each test subject’s performance estimations. Remember, benchmark test code typically has no real-world value, so “performance estimations” are about all I can promise here. Your mileage will vary. As a wise person once said:

“All this benchmarking is doing is proving what we already know: More code takes longer to execute.” – Ben Bangert (dev lead of Pylons)

What you should know about Round 3:

• The hardware/software platform is the same as in Round 1.
• Python v2.5.4 was used for all the Python tests. Ruby v1.8.6 was used for the Rails test.
• Apache v2.2.3 was used, with mod_wsgi v2.5 for the Python tests and Phusion Passenger v2.2.5 for the Rails test.
• Note that while the mod_wsgi tests in Round 1 were using “daemon mode”, the mod_wsgi tests here are using “embedded mode” which is often faster (although not necessarily the recommended configuration).
• The Python component versions used here were: SQLAlchemy v0.5.6, Genshi v0.5.1, Mako v0.2.5, and Jinja 2.2.1.
• ApacheBench was run with -n 10000 -c 10.
• Each test was run several times to make sure that there were no anomalies, with the “optimum average” chosen as the numbers represented here.
• In an attempt to try to make these tests more scientific, I am now including the source code for each test.

Plain WSGI – wsgi_test_src.zip

Let’s start with a basic “hello world” WSGI test to set the baseline:

The 'hello world' test:
 
WSGIServer/0.1
Document Path:          /
Document Length:        12 bytes
Requests per second:    1532.76 [#/sec] (mean)
 
 
Apache/2.2.3 (mod_wsgi)
Document Path:          /
Document Length:        12 bytes
Requests per second:    5549.73 [#/sec] (mean)

As you can see, mod_wsgi allows Apache to put some serious muscle into Python’s WSGI. It’s no wonder that mod_wsgi is quickly becoming the de-facto standard for Apache Python deployment.

TurboGears v2.0.3 – tg2_test_src.zip

As I mentioned in Round 1, TurboGears has quickly become my web framework of choice. Building a “best-of-breed” component stack on top of Pylons may not be easy, but when it works I believe it pays off immensely (e.g. Ubuntu).

The 'hello world' test:
 
Document Path:          /
Document Length:        12 bytes
Requests per second:    935.13 [#/sec] (mean)
 
 
Genshi template test:
 
Document Path:          /genshi_hello/
Document Length:        923 bytes
Requests per second:    601.07 [#/sec] (mean)
 
 
Mako template test:
 
Document Path:          /mako_hello/
Document Length:        932 bytes
Requests per second:    723.18 [#/sec] (mean)
 
 
Jinja2 template test:
 
Document Path:          /jinja_hello/
Document Length:        937 bytes
Requests per second:    764.35 [#/sec] (mean)
 
 
Database query tests:
 
Document Path:          /raw_sql/
Document Length:        1270 bytes
Requests per second:    569.71 [#/sec] (mean)
 
Document Path:          /genshi_sql/
Document Length:        2409 bytes
Requests per second:    388.96 [#/sec] (mean)
 
Document Path:          /mako_sql/
Document Length:        2418 bytes
Requests per second:    515.07 [#/sec] (mean)
 
Document Path:          /jinja_sql/
Document Length:        2472 bytes
Requests per second:    535.06 [#/sec] (mean)

TurboGears v2.1a1 – tg21_test_src.zip

The TurboGears community is alive and well, and just before finishing this post the release of v2.1 Alpha 1 was announced. Significant improvements were made to TG’s object dispatch, so I was excited to see what the new numbers would look like.

The 'hello world' test:
 
Document Path:          /
Document Length:        11 bytes
Requests per second:    1118.92 [#/sec] (mean)
 
 
Genshi template test:
 
Document Path:          /genshi_hello/
Document Length:        923 bytes
Requests per second:    713.97 [#/sec] (mean)
 
 
Mako template test:
 
Document Path:          /mako_hello/
Document Length:        932 bytes
Requests per second:    812.64 [#/sec] (mean)
 
 
Jinja2 template test:
 
Document Path:          /jinja_hello/
Document Length:        937 bytes
Requests per second:    1000.98 [#/sec] (mean)
 
 
Database query tests:
 
Document Path:          /raw_sql/
Document Length:        1270 bytes
Requests per second:    673.04 [#/sec] (mean)
 
Document Path:          /genshi_sql/
Document Length:        2409 bytes
Requests per second:    433.45 [#/sec] (mean)
 
Document Path:          /mako_sql/
Document Length:        2418 bytes
Requests per second:    601.89 [#/sec] (mean)
 
Document Path:          /jinja_sql/
Document Length:        2472 bytes
Requests per second:    630.66 [#/sec] (mean)

As you can see, the future of TurboGears looks very bright, and I only expect these numbers to get better as the final 2.1 release approaches (expected in Q1 2010).

Pylons v0.9.7 – pylons_test_src.zip

Pylons is the foundation which TurboGears is built upon, and for those who don’t need all “the extras” it is an excellent choice.

The 'hello world' test:
 
Document Path:          /hello/index
Document Length:        12 bytes
Requests per second:    2593.47 [#/sec] (mean)
 
 
Mako template test:
 
Document Path:          /hello/hello
Document Length:        932 bytes
Requests per second:    1737.03 [#/sec] (mean)
 
 
Database query tests:
 
Document Path:          /hello/raw_sql
Document Length:        1270 bytes
Requests per second:    964.07 [#/sec] (mean)
 
Document Path:          /hello/hellodb
Document Length:        2418 bytes
Requests per second:    831.26 [#/sec] (mean)

Django v1.1 – django_test_src.zip

While Django is probably Python’s most popular web framework, I have never really clicked with it. Its way of doing things is significantly different than TG/Pylons, but if you don’t mind its “MTV” architectural pattern and can survive without SQLAlchemy (although there are hacks), it is a very capable framework.

The 'hello world' test:
 
Document Path:          /hello
Document Length:        12 bytes
Requests per second:    3376.48 [#/sec] (mean)
 
 
Django template test:
 
Document Path:          /hellos
Document Length:        936 bytes
Requests per second:    1781.43 [#/sec] (mean)
 
 
Database query test:
 
Document Path:          /hellodb
Document Length:        2476 bytes
Requests per second:    972.11 [#/sec] (mean)

Bottle 0.5.8 – bottle_test_src.zip

Bottle caught my attention a few months ago while surveying the WSGI landscape for an alternative to PHP when building websites that only require a few pieces of dynamic functionality. As a “micro-framework”, it does an excellent job at bridging the gap between “pure WSGI” and its “feature-rich” big brothers.

The 'hello world' test:
 
Document Path:          /
Document Length:        12 bytes
Requests per second:    5545.96 [#/sec] (mean)
 
 
Mako template test:
 
Document Path:          /hello
Document Length:        923 bytes
Requests per second:    3588.04 [#/sec] (mean)
 
 
Database query test (SQLAlchemy):
 
Document Path:          /hellodb
Document Length:        2413 bytes
Requests per second:    1479.54 [#/sec] (mean)

As you can see, there was practically no difference in speed between Bottle and pure WSGI in a basic “hello world” test. Even with the addition of Mako and SQLAlchemy, Bottle performed significantly faster than a bare Pylons or Django setup.

(Note: Mako template inheritance was not used in the Bottle Mako test.)

Ruby on Rails v2.3.3 – rails_test_src.zip

What web technology shootout would be complete without Rails? Although I am not a Rails fan (mostly because I can’t stand Ruby), I believe that in many ways we owe much of the popularity and success of the web framework movement to Rails.

The 'hello world' test:
 
Document Path:          /hello/hello
Document Length:        12 bytes
Requests per second:    1048.18 [#/sec] (mean)
 
 
Rails template test:
 
Document Path:          /hello/hellos
Document Length:        937 bytes
Requests per second:    949.54 [#/sec] (mean)
 
 
Database query test:
 
Document Path:          /hello/hellodb
Document Length:        2482 bytes
Requests per second:    734.18 [#/sec] (mean)

(Note: I think my version of Ruby was using a newer version of the SQLite3 library than my version of Python. This may or may not have given Rails an advantage on the database query test.)

For those of you who like charts, here’s a comparison of the “return hello world” test results for each framework:

The return hello world test chart

And here’s a comparison of the template test:

The template test chart

Finally, here’s the template test with the database query added:

Template test with database query

Closing thoughts

• Gone are the days of Rails as “the one framework to rule them all.” C’mon guys, it’s not 2005 anymore.
• Python has gotten a lot of flak from other language camps for its abundance of web frameworks. While some Pythonists have been embarrassed by this, I only see it as proof of Python’s success as a language. My retort would be: “If it was as easy to write a web framework in your language as it is in Python, you’d have the same problem!”

Coming soon: “Round 4: Micro-frameworks — Quick & Dirty, and leaving PHP in the dust!”

(Note: Please don’t leave benchmarking requests as comments. If you’d like me to take a look at your favorite framework, direct message me on twitter. Thanks!)

Recently I went on a benchmarking spree and decided to throw ApacheBench at a bunch of the different web development technology platforms I interact with on a day-to-day basis. The results were interesting enough to me that I decided I’d take a post to share them here.

Disclaimer: The following test results should be taken with a massive grain of salt. If you know anything about benchmarking, you will know that the slightest adjustments have the potential to change things drastically. While I have tried to perform each test as fairly and accurately as possible, it would be foolish to consider these results as scientific in any way. It should also be noted that my goal here was not to see how fast each technology performs at its most optimized configuration, but rather what a minimal out-of-the-box experience looks like.

Test platform info:

• The hardware was an Intel Core2Quad Q9300, 2.5Ghz, 6MB Cache, 1333FSB, 2GB DDR RAM.
• The OS was CentOS v5.3 32-bit with a standard Apache Webserver setup.
• ApacheBench was used with only the -n and -c flags (1000 requests for the PHP frameworks, 5000 requests for everything else).
• Each ApacheBench test was run 5-10 times, with the “optimum average” chosen as the numbers represented here.
• The PHP tests were done using the standard Apache PHP module.
• The mod_wsgi tests were done in daemon mode set to 2 processes/15 threads.
• The SQL tests were done with mysqli ($mysql->query()) on PHP, and SQLAlchemy (conn.execute()) on Python fetching and printing 5 rows of data from a sample database.

Apache v2.2.3

We will start with the raw Apache benchmark.

For this test, Apache loaded a simple HTML file with random text:

Document Length:        6537 bytes
Requests per second:    8356.23 [#/sec] (mean)

As expected, Apache is lightning fast.

Ok, so now that we’ve set the high water mark, let’s take a look at some popular web technology platforms…

PHP v5.2.8

For this test, I created a PHP file that simply printed the $_SERVER array and some random text:

Document Length:        6135 bytes
Requests per second:    2812.22 [#/sec] (mean)

Next, I decided to add some mysqli() code to connect to a database and print out a few rows of data:

Document Length:        283 bytes
Requests per second:    2135.46 [#/sec] (mean)

Not at all bad, but who really wants to build a webapp using plain old PHP anymore?

Next I decided to test the two PHP frameworks that I am most familiar with.

CakePHP v1.2.5

For this test, I did a basic CakePHP install, and added a route that printed “hello world” from a template:

Document Length:        944 bytes
Requests per second:    42.00 [#/sec] (mean)

As you can see, frameworks can be a blessing and a curse. On a side note, CakePHP was the only test that gave me failed requests.

Kohana v2.3.4

For this test, I did a basic install and loaded Kohana’s default welcome/index setup (which seemed lightweight enough to me to qualify as the test focal point):

Document Length:        1987 bytes
Requests per second:    100.11 [#/sec] (mean)

Kohana fared much better than CakePHP, but it appears as though PHP’s speed breaks down quickly once you start building a framework around it.

Python v2.5.4

Unlike PHP, Python wasn’t specifically created for the web. However, several years ago the WSGI spec was introduced which makes building web applications with Python a breeze. There are many different ways to depoly a WSGI app, but my tests were limited to Python’s built-in WSGIServer, and Apache’s mod_wsgi.

WSGIServer

For this test, I used WSGI.org’s test.wsgi:

Document Length:        4123 bytes
Requests per second:    1658.35 [#/sec] (mean)

These results were quite impressive to me, considering that this is pure Python at work.

mod_wsgi

mod_wsgi is Apache’s module to process WSGI scripts. This test used the exact same script as the previous WSGIServer example:

Document Length:        9552 bytes
Requests per second:    3994.30 [#/sec] (mean)

Don’t look now, but those results are better than the vanilla PHP test!

Next, I added SQLAlchemy imports to the test file and did a conn.execute(), in an attempt to try to mimic the PHP mysqli() test:

Document Length:        10212 bytes
Requests per second:    3379.87 [#/sec] (mean)

I expected to lose quite a bit of speed with the introduction of an ORM, but it looks like SQLAlchemy is quite fast. These results were still over 1,000reqs/s more than the PHP example.

TornadoServer

There’s been a lot of buzz lately about the new TornadoServer, so I thought I’d throw it in the mix. This test uses their basic “hello world” example:

Document Length:        12 bytes
Requests per second:    3330.00 [#/sec] (mean)

Nothing shocking here at first glance, but the Tornado is built to handle extreme loads. So, I increased the concurrency to 1,000 and threw 10,000 requests at it:

Document Length:        12 bytes
Concurrency Level:      1000
COMPLETE requests:      10000
Requests per second:    3288.15 [#/sec] (mean)

Almost exactly the same results! On a side note, none of the other tests subjects were able to handle a concurrency of 1,000.

TurboGears v2.0.3 (mod_wsgi)

TurboGears has quickly become my framework of choice, so naturally it was the first Python framework I tested.

For this test I setup a default quickstart project, and then added a hello() controller method which simply returned a “hello world” string:

Document Length:        11 bytes
Requests per second:    470.18 [#/sec] (mean)

Next, I added a simple Genshi template with 1 level of inheritance:

Document Length:        1351 bytes
Requests per second:    143.26 [#/sec] (mean)

While those results are not terrible, they should encourage you to check out the other templating options.

Next, I performed the same test with Mako:

Document Length:        1534 bytes
Requests per second:    394.41 [#/sec] (mean)

As you can see, Mako is significantly faster than Genshi.

TurboGears also supports Jinja2, so I thought I’d also see how it held up:

Document Length:        1020 bytes
Requests per second:    401.04 [#/sec] (mean)

I was not expecting Jinja2 to be faster than Mako, so I am not entirely sure about these results. However, the difference is only slight, and might have something to do with the fact that the dotted template names finder is disabled for Jinja templates.

Pylons v0.9.7 (mod_wsgi)

Mark Ramm (the current development lead of TurboGears) has been quoted as saying “TurboGears is to Pylons as Ubuntu is to Debian.” With that in mind, I naturally wanted to see how much speed I was losing with TurboGears’ “additions”.

For this test I tried to mimic the first TurboGears test (basic layout, returning a string):

Document Length:        11 bytes
Requests per second:    1961.00 [#/sec] (mean)

While I was expecting a gain from TurboGears’ omissions, I wasn’t quite expecting over four times the increase. I can only assume that the bare Pylons setup is pretty minimal, while TurboGears bootstraps a lot of stuff on for you by default.

Duplicating the Mako test in Pylons was also quite fast:

Document Length:        9743 bytes
Requests per second:    1183.39 [#/sec] (mean)

Bottle v0.5.7 (mod_wsgi)

Bottle caught my attention recently as a minimalistic Python web framework, so I thought I’d put it through the grind as well.

For this test, I used their simple “hello world” example:

Document Length:        12 bytes
Requests per second:    3912.43 [#/sec] (mean)

As you can see, Bottle is very close to bare WSGI as far as speed goes. On a side note, adding a sample template using Bottle’s default templating package didn’t seem to change these numbers at all.

Next I tested Bottle with its templating system switched to Mako:

Document Length:        41 bytes
Requests per second:    3134.61 [#/sec] (mean)

Then I added a SQLAlchemy and performed a query:

Document Length:        9374 bytes
Requests per second:    2503.27 [#/sec] (mean)

What was exciting to me about this test was that I found myself looking at a minimal Python framework equipped with a full templating system and ORM which all performed faster than the PHP+mysqli() test.

Django v1.1 (mod_wsgi)

I don’t use Django, but I thought it would be worth including in this report. This is a basic Django setup printing out a simple hello world template:

Document Length:        1471 bytes
Requests per second:    564.41 [#/sec] (mean)

As expected, a minimal Django setup is quite fast. (Note: This test included a SQLite database query.)

Ruby On Rails v2.2.3 (mod_passenger)

I also don’t use Rails, but I figured this report would be incomplete without Rails on the list. So, I took MediaTemple’s Rails test app and loaded up a test page:

Document Length:        1291 bytes
Requests per second:    433.57 [#/sec] (mean)

Note that the test app used here is doing a little bit more than just returning “Hello World”.

My Conclusions

As I said in the disclaimer, to arrive at any sort of decisive conclusions based on these numbers alone would be foolish. However, here is what all of this left me thinking:

• All of these technologies performed adequately for use with most projects.
• CakePHP was the only test that produced less than 100 requests per second. However, there are many articles available on optimizing CakePHP.
• These results made me wish I had moved away from PHP much earlier in my web development carrier. There are just better options these days when it comes to choosing a web development language (imho). If you must use PHP for a large project, I strongly suggest that you look into an accelerator.
• The Bottle+Mako+SQLAlchemy results struck me as an extremely minimal drop-in replacement for simple PHP sites.
• mod_wsgi is a huge winner in deploying Python webapps. Way to go Graham!
• Do your own bencharking, because your mileage will vary!

Like what you’ve seen here? Then why not check out Round 2?

(Note: Please don’t leave benchmarking requests as comments. If you’d like me to take a look at your favorite framework, direct message me on twitter. Thanks!)