When running performance tests, we'll be using specific thread configurations depending on our objectives. But, the real question is, what inputs should those thread configurations use. Do I send 1000 users or just 50? How many requests should there be to really test the website’s performance?
Worry not, there is a way to find out. The magic word for this is the baseline. Yeah, we need to find the baseline for an app. For example, we have an app in production. For choosing the numbers that we're going to use in our performance tests (spike/load/stress), we'll take as baseline the production env statistics from Grafana (can be any log service).
To find the baseline, let’s follow the next example:
- Go to grafana.
- TimeRange: Last 7 Days.
- Scroll down to the
Client Workloadgraphs. - Focus on the
Incoming Requests By Source And Response Code. - Find the highest point in the graph.
- Zoom in.
This way you have a starting point, from where you can decide the stages you want to put your app through during performance testing. You know that in the last 7 days you had a max of 187 ops/s. You can go from here, just ask yourself questions like, how fast can the system recover, how much can it maintain in this state and so on. Sure you can try 1 milion users, but will that ever be close to reality of your userbase?
