Predicting Failures

Why Prediction Matters

Most performance tests are run at a fixed load level. For example, the load is set at 100 concurrent users. The test either passes or fails at that load. But this binary view misses something important: how close to the edge is the system?

If your system barely passes at 100 users but would collapse at 110, that is very different from a system that handles 100 users comfortably and shows no degradation up to 300. Both “pass” in a simple pass/fail test, but one is a ticking clock.