Network Issues

For load and stress tests, the network connection between the test client and the server is critical. For the connection between the server and the test client you must provide the full bandwidth that an equal number of real users would use when accessing your server!

This means that you obviously can not conduct a serious load test with 500 users requesting a 5 MB file over a single 56kb modem connection.

Additionally, if your are running the test from a remote location, the number and the performance of the hops (router/firewalls etc.) can influence the test. The optimum testing environment is to  run the server and client within the same networked environment (i.e., on the same LAN).

For heavy load testing, it is the best to connect both the client and server to a high performance network switch. Since Webserver Stress Tool on a fast PC can easily work with more bandwidth than a 100 Mbit LAN can deliver, even a Gigabit Ethernet may be a good idea.

For tests over internet connections like T1, DSL etc. you have to make sure that the amount of data created by your tests does not exceed the bandwidth of these connections. Use a bandwidth monitoring software like PRTG Traffic Grapher to monitor the bandwidth usage (www.paessler.com/prtg)

Usually for performance and smaller load tests a leased line with 500 kb/s or more should be enough, but more bandwidth will always give you more reliable results. Furthermore, you have to make sure that the “travel time” of the data is far below the request times you measure. Otherwise, measured values will be unreliable.

Everything below a 300 kb/s connection should be considered vague testing, although it can give good results under some circumstances, e.g. for long running web server scripts that only produce very little HTML code. The same applies for modem connections.