Types of performance test
| Type | Question it answers |
|---|---|
| Load | Does it hold at the expected peak (for example 500 concurrent users)? |
| Stress | Where does it break, and how does it fail? |
| Spike | Can it survive a sudden surge (a campaign, a TV mention)? |
| Soak | Does it degrade over hours (memory leaks)? |
| Scalability | Does adding servers actually add capacity? |
Metrics that matter
- p95 and p99 response times, not just averages
- Error rate under load
- Throughput (requests per second)
- Database CPU, connections and slow queries
- Cost per 1,000 requests, especially with paid AI APIs
A simple plan
- Define realistic user journeys and peak numbers from analytics.
- Set pass/fail targets (for example p95 under 800 ms, errors under 1%).
- Script the journeys in k6 or JMeter.
- Run baseline, then ramp to peak, then stress.
- Fix the top bottleneck, re-run, repeat.
Web speed is a performance test too
For public sites, Core Web Vitals (LCP, INP, CLS) affect user experience and SEO. We tune images, scripts and caching on every SEO-ready website.
Example test plan for a booking app
| Item | Value |
|---|---|
| Peak expectation | 300 concurrent users after a campaign |
| Journeys | Browse availability (60%), book (25%), view booking (15%) |
| Targets | p95 under 800 ms; errors under 1%; CPU under 70% |
| Phases | 5-minute ramp, 15-minute steady state, 5-minute stress to 2× peak |
| Typical fixes | Add database indexes, cache availability, queue emails, compress images |
Reading the results
- A rising p95 with flat CPU often points to database locks or slow third-party calls.
- Errors that appear only at high load usually indicate connection limits or timeouts.
- Always retest after each fix; one bottleneck often hides the next.
Load-test before launch From $50/h
Tell us expected traffic. We design and run the test.
Frequently asked questions
How many users should I test for?
Use peak concurrent users you expect in the next 6–12 months, then test at 2–3 times that for headroom.
Can you test apps that call LLM APIs?
Yes, and we also test rate limits and costs, since LLM calls can dominate both latency and budget.
Get a free quote in 24 hours
Tell us what you need. We reply with scope, timeline and a fixed price.