Apache2.2 + Mongrel: what do you think about these perfs?

Rails
1

Hi all,

I’ve been spending quite a lot of time trying to install a decent RoR
server on my dedicated server (Ubuntu 6.06 LTS), and now, everything
works.

However, the performances are not really what I had expected… I would
like to know what you think about it.

Here is my config: 2GHz VIA proc, 1Gb RAM, SATA-II HD.

I have apt-got ruby 1.8.4, mysql5 and installed rails 1.1.4 by means of
rubygems (0.9.0). Then I installed the mysql extension (2.7}, mongrel
(0.3.13.3) and mongrel cluster (0.2.0) and uploaded my app.

Here is my mongrel_cluster.yml:

port: “8000”
environment: production
pid_file: log/mongrel.pid
servers: 3

Then, I compiled Apache2.2. Here are my configure options:

./configure --prefix=/usr/local/apache2 --enable-modules=‘alias asis
auth_basic auth_digest authn_file authz_user autoindex access cgi cgid
charset_lite dir env http imagemap include log_config mime negotiation
setenvif status deflate info proxy proxy_balancer rewrite so headers’

Not too much, and static only… I don’t see how I could improve this…

Now, here is my vhost config:

<VirtualHost *>
ServerName dedibox
DocumentRoot /home/noe/cukv3/public
<Directory “/home/noe/cukv3/public”>
Options FollowSymLinks
AllowOverride None
Order allow,deny
Allow from all

<Proxy balancer://cukv3>
BalancerMember http://127.0.0.1:8000
BalancerMember http://127.0.0.1:8001
BalancerMember http://127.0.0.1:8002

ProxyPass /images !
ProxyPass /stylesheets !
ProxyPass /javascripts !
ProxyPass / balancer://cukv3/
ProxyPassReverse / balancer://cukv3/

The whole thing works, however, as I have said, I am not satisfied by
the performances… Here is what I get for a middle-complexity action,
which returns 10 articles from a given category (n-to-n relationship).
(In fact, mysql’s CPU usage never gets > 3%, so it is clearly ruby that
slows it down…)

These benches have been realized with apache bench, from the box itself.
So there is no delay due to the network. I know httperf is better, but I
couldn’t manage to get it to work, because it took too much CPU.

So, here are the results:

  1. Direct connexion to a mongrel server:

    1.1) Dynamic

    ab -n 100 -c 10 127.0.0.1:8000/articles/list?category_id=2
    Result: 5.70 #/s

    1.2) Static

    ab -n 10000 -c 10 127.0.0.1:8000/stylesheets/img/logo.gif
    Result: 439 #/s

  2. Through Apache2.2 and mod_proxy_balancer

    2.1) Dynamic

    ab -n 100 -c 10 127.0.0.1/articles/list?category_id=2
    Result: 6.38 #/s

    2.2) Static

    ab -n 100 -c 10 127.0.0.1/stylesheets/img/logo.gif
    Result: 1703 #/s

Which seems to be unsufficient for production use…

So, what do you think? Is it normal? Did I do something wrong? In fact,
I would be interested in points of comparison, and thoughts about these
results.

Thank you very much!

Nauhaie

2

On Mon, 2006-07-17 at 00:09 +0200, Nauhaie wrote:

Everything looks fine from first glance.

> > 2.2) Static > > ab -n 100 -c 10 127.0.0.1/stylesheets/img/logo.gif > Result: 1703 #/s

You should do one more test:

  1. Create a small controller named “test_controller.rb” and put an index
    action in it that render_text “test” as the only thing it does.
  2. Re-run your above test and see what kind of performance you get.

This is most likely your max Rails performance, if other pages aren’t at
this level then you have to work on the Rails side to get them faster.
If this measurement is the same as your /articles test then you’ve got
something configured wrong since it has to be faster.

Check out ruby-prof as a way to get good performance tuning measurements
for that specific action:

http://rubyforge.org/projects/ruby-prof

Specifically the part about using the ruby-prof library to measure one
function.

–
Zed A. Shaw

http://mongrel.rubyforge.org/
http://www.railsmachine.com/ – Need Mongrel support?

3

Just as a sanity check, I have an under-development app that I switched
over
to production mode on a MacBook Pro (not the fastest server on earth)
and
httperf shows a single Mongrel serving up somewhere around 27 req/s for
dynamic pages. Remember, this is a laptop I’m benchmarking. Your
server
should outpunch it by a significant amount.

Nauhaie wrote:

ab -n 100 -c 10 127.0.0.1:8000/articles/list?category_id=2
Result: 5.70 #/s

–
View this message in context:
http://www.nabble.com/Apache2.2-%2B-Mongrel%3A-what-do-you-think-about-these-perfs--tf1952024.html#a5353289
Sent from the RubyOnRails Users forum at Nabble.com.

4

Hi!

  1. Create a small controller named “test_controller.rb” and put an index
    action in it that render_text “test” as the only thing it does.
  2. Re-run your above test and see what kind of performance you get.

You are right, I should probably have done that in the first place…
So, here are my results:

Direct to Mongrel:

ab -n 100 -c 10 127.0.0.1:8000/test ==> 45#/s

Through Apache to 3 Mongrels:

ab -n 100 -c 10 127.0.0.1/test ==> 45#/s, too

Does it seem you right?

Thank you!

Nauhaie

5

If it might help, I have tried it with the same 3 Mongrels and Pound
instead of Apache2.2 + mod_proxy_balancer.

I get 6.14 #/s (I had 6.38 with last config) with my script. With the
simple Hello World, I get 40.86 (45 with last config). It appears that
Pound is no better than Apache2.2… Can someone confirm that please?

I would really be interested in what performances you guys get, because
I have pretty much nothing to compare with…

Thank you!

Nauhaie

6

Ok thank you. I will try to see what I can do. At least, I am glad my
configuration is ok :wink:

Nauhaie

7

On Mon, 2006-07-17 at 23:10 +0200, Nauhaie wrote:

If it might help, I have tried it with the same 3 Mongrels and Pound
instead of Apache2.2 + mod_proxy_balancer.

I get 6.14 #/s (I had 6.38 with last config) with my script. With the
simple Hello World, I get 40.86 (45 with last config). It appears that
Pound is no better than Apache2.2… Can someone confirm that please?

I would really be interested in what performances you guys get, because
I have pretty much nothing to compare with…

Yep, basically you’re putting your efforts in the wrong place. If you
run a test controller and you get 40.86, and then your page gets 6.14,
then changing out the balancer will do absolutely nothing for you.
You’ve got to work on making your rails actions faster. Nothing can
save you. You’re doomed. Time to work on the real problem. :slight_smile:

Check out ruby-prof for some hints, also look at some earlier posts by
Stefan K. and Scott B. for other tools. Otherwise, just keep
benching and tweaking until you find out the problem. Your
production.log is the key to your problem.

–
Zed A. Shaw

http://mongrel.rubyforge.org/
http://www.railsmachine.com/ – Need Mongrel support?