How many Thin server instances are best ?

I wondered about how many parallel thin you need to get the optimal throughput. I often saw values from 3 to 10 servers in different configuration examples around the net. Time to get some real numbers.

To get a meaningful benchmark i took an old dual 1,3 GHz P3-s system with the latest ubuntu. I did install rails 2.0.2 and a testing project with it, then created a very simple controller, that just passes a “hello world” string to the view. Also, no database connections should be made.

The server is started with

rake thin:cluster:start RAILS_ENV=production SIZE=5 PORT=8000

You’ll need the latest thin 0.5.2 for it, as 0.5.1 has some serious bug that prevents it from running daemonized. The command above will start 5 thin instances, listening on ports 8000-8004. To actually benchmark the number of thin instances used, i did modify the upstream entries in nginx:

# only one server used in this example
upstream thin {
   server 127.0.0.1:8000;
   #server 127.0.0.1:8001;
   #server 127.0.0.1:8002;
   #server 127.0.0.1:8003;
   #server 127.0.0.1:8004;
}

The results are not really supprising and show that the nginx + thin combo scales well even up to 100 concurrent users. All Tests were done with

ab -n 10000 -c 20 http://localhost/foo/

where the “foo” controller was the minimal one returning the “hello world” page.

The results can be seen in this graph:

As you can see, the best results for this dual cpu machine was indeed running two instances of thin. What is interesting to see that thin+nginx scale quite well over many concurent requests.

Of course this is an idealized test with minimal load times for each pagecall. In the next version i’ll put in a random delay before the template is rendered to simulate real wold loading times and slow operations (users up/downloading large files, etc ..)