Before running performance tests, you need a shared vocabulary for how systems behave under load. Understanding basic terms like response time, throughput, latency and bottlenecks makes it easier for QA, developers and stakeholders to discuss results.
Key Performance Concepts
Response time is how long a system takes to handle a request, throughput is how many requests it can handle per second, and latency refers to the delay between sending a request and starting to receive a response. Bottlenecks are the components (for example database, CPU, network) that limit overall performance when demand increases.
Example performance targets:
- Homepage median response time < 500 ms at 200 requests/second
- Checkout error rate < 1% under expected peak load
- API P95 latency < 800 ms during business hours
- No CPU core constantly above 85% utilisation in steady stateWith these concepts in mind, you can design tests that measure the right things instead of just hitting the system with traffic.
Common Mistakes
Mistake 1 — Talking about performance without specifying load
This creates confusion.
❌ Wrong: Saying “the API is fast” without mentioning how many concurrent users were simulated.
✅ Correct: Always pair performance numbers with the conditions under which they were measured.
Mistake 2 — Ignoring percentile-based metrics
This hides user pain.
❌ Wrong: Looking only at average response time.
✅ Correct: Track P50, P95 or P99 latency to understand tail behaviour.