Finding the Bottleneck of Your Test Setup

When running load tests on a webserver you must make sure that you do not hit a performance limitation of your test client or your network.

The best way to find these limits is to run a ramp test with twice or three times the load you want to test with (or even more) and inspect the Test Client’s Health graph afterwards.

The graph for Network Traffic and local CPU Usage should ramp up with the increasing number of users. When either one hits a plateau you have found your limit - or the limit of the server.

E.g. if you are using a 100 Mbit network you may see the Network Traffic graph hitting the 100 Mbit/s bandwidth limitation of your network hardware.

To distinguish between client/network and server issues it is a good idea to also monitor the CPU Load/Network of the server which will also help to find out what the bottleneck is. If Webserver Stress Tool already indicates a limit, but your server is more or less idle, you need a machine with more testing power.

Also keep any eye on the Protocol Times graph. Under heavy loads sometimes the Time for local socket can rise sharply (above 10-50 ms) which also indicates a performance bottleneck.