Article 05

Performance testing: from business questions to decision-ready results

Plan workload, scenarios, data, environment, thresholds and observability so results show what to fix first.

16 min Tech Leads, developers, QA and platform teams 2026-07-17
Overview infographic
Threshold

Load

250 RPS

p95

780 ms

Errors

0.6%

Example situation

Start with the people using the system

The test reached 1,000 users, but is the system ready?

A high user count means little when the test does not match behavior, lacks pass/fail criteria and misses API, database and infrastructure metrics. A useful test answers a business question, such as whether booking completes during opening traffic.

Summary for business and everyday users

Performance testing: from business questions to decision-ready results

01

Start with the test reason: validate normal traffic, find limits or prepare for a campaign.

02

Define thresholds before running, not after seeing results.

03

Align application and infrastructure metrics with the load timeline.

Interactive explanation

See how the stages connect before reading the detail

Select a stage to understand what happens, what evidence to inspect and how one decision affects the next stage.

Step 01 / 04

01 / Define the question

Choose test type from what you need to learn

Smoke tests verify scripts; average-load tests validate normal use; stress tests exceed normal demand; spike tests model sudden jumps; soak tests reveal accumulation; breakpoint tests find limits. One type cannot answer every question.

Evidence to inspect

  • Which flow must not fail?
  • What are normal and peak loads?
  • Are we validating or finding a limit?
  • Can test data affect production?

In-depth explanation

Work through each issue in real operating context

Each section connects business impact with what a Tech Lead needs to inspect, including examples, evidence and constraints.

01 / Define the question

01

Choose test type from what you need to learn

Smoke tests verify scripts; average-load tests validate normal use; stress tests exceed normal demand; spike tests model sudden jumps; soak tests reveal accumulation; breakpoint tests find limits. One type cannot answer every question.

  • Which flow must not fail?
  • What are normal and peak loads?
  • Are we validating or finding a limit?
  • Can test data affect production?

02 / Build scenarios

02

Model behavior mix, not one endpoint

Real systems mix login, browse, search, save, reports and background APIs. Model observed proportions, think time and data that does not distort cache or database behavior.

Detail for Tech Leads

Tag flows and endpoints so p95 and error rate are visible per group; do not let aggregate averages hide critical transactions.

03 / Pass or fail

03

Thresholds connect experience and risk

Define response-time percentiles, error rate and transaction success before testing. p95 represents most users without being hidden by very fast requests, but targets must come from business context rather than copied examples.

Example threshold

A pattern, not a universal standard

Search: p95 below 800ms and errors below 1%; order confirmation: success above 99.5% with no duplicate transactions.

04 / Read the result

04

Place load beside metrics from every layer

When latency rises, align it with CPU, memory, event loop, connection pool, slow queries, locks, cache hit, queue depth and external APIs. Load-generator results alone show slowness but not its cause.

Checklist before action

01

Define the test reason

02

Build workload from evidence

03

Prepare test data

04

Set thresholds in advance

05

Collect every-layer metrics

06

Document conclusions and limits

References

Figures and examples create a discussion framework. Validate them against the real system and its constraints before deciding.

Reviewed by
SIS Engineering
Last reviewed
2026-07-17

Not sure where to start the review?

Use a preliminary tool or share the system context with SIS so the highest-priority work can be identified.

Back to Knowledge Base