How Many Mongrel Instances Should I Run?

There is no set number that is "best" since that depends on factors like the type of application, server hardware, how dynamic the appication is, etc.

I've found that 8-12 mongrel processes per CPU is about right, but I determined this by starting with 1 and then doing the following:

Baseline Your Server

Pick a URL to a small file that is running on your apache server and is not served by Mongrel at all. This URL will be your "best possible baseline".

Build your baseline measurement first. Using httperf, measure the speed of your URL so that you know how fast you could possibly get if you served everything static in ideal situations.

Make sure you do this on a different machine over an ideal network. Not your damn wifi over a phone line through sixteen poorly configured routers. Right next to the box your testing with a fast switch and only one hop is the best test situation. This removes network latency from your test as a confounding factor.

Baseline Rails and Mongrel

Pick a page that's a good representative page for your application. Make sure you disable logins to make this test easier to run. Hit this Rails page and compare it to your baseline page.

  • If your rails measurement is faster than your baseline measurement then you screwed up. Rails shouldn't be faster than a file off your static server. Check your config.
  • If your rails measurement is horribly slow compared to baseline then you've got some config to do before you even start tuning the number of process. Repeat this test until one mongrel is as fast as possible.

Tweak

Once you've got a Rails page going at a reasonable speed, then you'll want to increase the --rate setting to make sure that it can handle the reported rate.

Finally, alternate between adding a mongrel process and running the test with the next highest rate you'd get. Stop when adding one more server doesn't improve your rate.

Make sure you run one round of the test to get the server "warmed up", and then run the real one. Heck, run like 5 or 6 just to make sure you're not getting a possibly bad reading.

Example

  1. Run the test and find out that one mongrel can support a --rate of 120 req/second.
  2. Add another mongrel and run the test again with --rate 240. It handles this just find so you add another and get --rate 360.
  3. Try another one and it dies. Giving --rate 480 gets you only a rate of 100. Your server has hit it's max and broke.
  4. Try tuning the --rate down at this and see if it's totally busted (like, 4 mongrels only gets you --rate 380) or if it's pretty close to 480.
  5. That should do it. A good practice is to also look at the CPUs on the server with top and see what kind of thrashing you give the server.

HTTPERF

Here's the commands I use for each test, but read the man page for httperf so that you learn to use it. It's an important tool and just cut-pasting what I have here is not going to do it for you.

  • httperf --server www.theserver.com --port 80 --uri /tested --num-conns <10 second count>
  • httperf --server www.theserver.com --port 80 --uri /tested --num-conns <10 second count> --rate <reported req/sec>

Where <10 second count> is enough connections to make the test run for 10 seconds. Start off with like 100 and keep raising it until it goes for 10 seconds.

Where <reported req/sec> is whatever httperf said the estimated requests/second were. What you're doing here is seeing if it really can handle that much concurrency. Try raising it up and dropping it down to see the impact of performance on higher loads.

Have fun.