Capturing the upper limit of your web server’s resources can be simplified by at least one design decision that you’ve probably already made: using a load balancer. In order to confirm your ceiling estimates with live traffic, increase the production load carefully on some web servers and measure the effects it has on resources. Increase the load by pulling machines from the live pool of balanced servers, which increases the load on the remaining servers commensurately. I want to emphasize the importance of using real traffic instead of running a simulation or attempting to model your web server’s resources in a benchmark-like setting.
The Art of Capacity Planning: Scaling Web Resources John Allspaw