measuring performance - using real clicks vs "ab" command

views:

answers:

measuring performance - using real clicks vs "ab" command

I have a web site in closed beta, developed in Django, runs with Mysql on Debian.

In the last few days, the main page has been showing a slowdown. For every ten clicks, one or two receives extremely slow response (10 secs or more), others are as fast as they used to be.

When I was searching for the problem, I ran into this issue that I couldn't grasp:

top command shows that when I request the main page, mysql shoots up to 90% - 100% cpu usage. I get the page just as the cpu use gets back to normal. So, I thought, it is db.

Then I called ab with parameters -n 1000 -c 5, I got decent performance, about 100 pages per second, just as it was before the slowdown. I would imagine a worse performance as 10-20% of requests take 10 secs to load.

Is this conflict between ab and "real" clicks normal, or am I using ab in a wrong configuration?

ab doesn't execute many parts of the page (javascript i.e.) so you'll notice probably a sensible difference in pressure to the webserver.

Ass3mbler 2009-11-27 13:26:06

Thanks for the answer. However, the slowdown does not stem from the loading or executing javascript (js files are compressed, versioned and have far-future expiration days).

shanyu 2009-11-27 13:30:31

But does the javascript do some other call to the webserver (i.e. ajax) when executed on the client side?

Ass3mbler 2009-11-27 13:33:10

No, there's no ajax on the page I am benchmarking. It may be worth to mention that the browser freezes while waiting for the server (for about 10 secs), and then gets and renders the page very fast.

shanyu 2009-11-27 13:39:33

ok... so the problem must be on other page element not handled by ab, like session data. Do you store your session on mysql?

Ass3mbler 2009-11-27 15:03:53

ansaurus

tags:

views:

answers:

measuring performance - using real clicks vs "ab" command

related questions