Question

I’m going to perform a load and stress test on a webpage using Apache JMeter, but I’m not very sure about the appropriate network setting. Is it better to connect the two machines, the server with the webpage and the client running JMeter via local network or via the internet. Using the internet would be closer to the real scenario, but with a local network the connection is much more stable and you have more bandwidth for more requests and the same time.
I’m very thankful for opinions!

Answer 1

These are in fact two styles or approaches to load testing, both are valid.

The first you might call Lab testing. Where you minimise the number of factors that can affect throughput/resp. times and really focus the test on the system itself.

The second is the more realistic scenario where you are trying to get as much coverage as possible by routing requests through as many of the actual network layers that will exist when the system goes live.

The benefit of method 1 is that you simplify the test which makes understanding and finding any problems much easier. The problem is you lack complete coverage.

The benefit of method 2 is that it is not only more realistic but it also gives a higher level of confidence – esp. with higher volume tests, you might find you have a problem with a Switch or Firewall and it is only with this type of testing that you identify such issues. The problem is it can make finding any issues harder.

So, in short, you really want to do both types. You might find it easier to start with the full end to end test from outside in, and then only move to a more focused test if you find that you need to isolate / investigate a problem. That way you stand a chance of reducing the amount of setup work whilst still getting the maximum benefit from testing.

Note: Outside in means just that, your test rig should be located outside of the LAN (assuming this is how live traffic will flow). These days this is easy to setup using cloud based hardware.

Also note: If the machine that you are running the tests from is the same in both cases then routing the traffic via the internet (out of your LAN and then back in again) is probably not going to tell you anything useful and could actually cause a false negative in your results (not to mention network problems for your company!)