Performance Testing vs Load Testing Explained

Performance testing is the broader discipline and load testing is one area inside it. See what each measures, with stress, soak and spike in one table.

Mustafa BayramogluMustafa BayramogluAbstract illustration of one steady amber line beside a second line that fans out and frays at the edge

Performance testing is the broader discipline, and load testing is one area inside it. A load test asks how the system performs under expected normal conditions. Other tests, such as stress, soak and spike, change the shape of the load to ask different questions. This page sets out what each one measures. For the full method of running a load test on an API, see the load testing guide.

What performance testing covers

Performance testing looks at an application’s speed, responsiveness and stability. Functional testing checks whether a feature works. Performance testing checks how well the system holds up under pressure.

Three dimensions come up in almost every definition:

  • Speed: how fast the application responds to a request.
  • Scalability: how well the system handles an increasing volume of users or data.
  • Stability: whether the system runs reliably over a long period without crashing or leaking memory.

Unlike functional tests, performance tests compare results to benchmarks. A functional test has a right answer. A performance test has a target you chose: a latency you can live with, an error rate you can tolerate, a ceiling you want to stay under.

Because the term covers so much, “we ran a performance test” says little on its own. The useful follow-up is always which question the test asked.

What load testing measures within it

Load testing is one area of performance testing. Its question is narrow: how does the system perform under expected normal conditions?

Take a shop API with three calls: log in, list products, place an order. A functional test sends each call once and checks the answer. A load test sends the same sequence from many virtual users at once, for several minutes, and watches what changes.

The test gives you three readings to track as load rises: latency (p95 is the usual one), error rate, and requests per second. One request that returns 200 in 80 ms says little about the same request when hundreds of users are in the middle of checkout. That gap between a single request and the same request under concurrency is what a load test exists to expose.

Performance testing vs load testing, side by side

Performance testing Load testing
Scope Speed, responsiveness and stability One area inside it
Question How well does the system hold up under pressure? How does it perform under expected normal conditions?
Load shape Depends on the type of test The load you expect
Typical readings Response time, throughput, error rate p95 latency, error rate, requests per second

The Testsigma comparison says performance testing covers system behavior under diverse conditions. The Abstracta comparison calls load testing an important area of performance testing.

In practice the two names overlap. Teams say “performance test” when they mean a load test, and the reverse. If you need a place to start, run a load test first. It gives you a baseline at the traffic you expect, and every other type is easier to read against that baseline. The post on load, stress, soak and spike tests compares the load shape each one uses.

Stress, soak, spike and the other types, in one table

Each type changes the shape of the load to ask a different question. The wording below follows the k6 test types guide.

Type What it asks
Smoke Does your script work, and does the system perform adequately under minimal load?
Load How does the system perform under expected normal conditions?
Stress How does the system perform at its limits, when load exceeds the expected average?
Soak Is the system reliable over extended periods?
Spike Does the system survive sudden, short and massive increases in activity?
Breakpoint Where are the capacity limits, found by raising load gradually?

Two more types turn up in performance testing guides. Volume testing floods the system with large amounts of data to test database and storage limits. Scalability testing checks whether the system can scale up or out as resources or user demands grow.

Treat the names loosely. A stress test for one application is an average-load test for another, and there is no consensus on what these types are called. Ask what load shape a team means before you compare their results with yours.

The numbers each test reports

Whatever the type, a test reports a small set of numbers.

  • Response time: the time it takes from sending a request to receiving a response.
  • Throughput: the number of transactions or requests a system can handle per second.
  • Error rate: the percentage of requests that fail or return errors.

Read p95 instead of the average. The average hides the slow tail, and the tail is what your unluckiest users experience. In a ramp, also compare the load you applied with the throughput you got. When the users keep climbing and the request rate flattens, the API has reached its ceiling, and anything beyond that point is queueing.

How many virtual users to start with is its own question, covered in choosing virtual users for an API.

How a test becomes a result: ramps, thresholds and abort rules

The load shape is only half of a test. A ramp has three phases: climb to a target, hold it long enough to read stable numbers, then drop back to zero. Jumping straight to full load makes it hard to tell what broke the API and at what level. A gradual climb shows you the point where latency bends.

How you describe the load matters too. With a fixed number of virtual users, a slower API means iterations take longer and fewer new ones start, so the request rate tapers off just when the system is under stress. An arrival rate is an open model, which decouples the start of new iterations from how long the last one took. Pick the model that matches the question: a number of people at once, or a request rate you want to hold.

Then decide what counts as a pass. Thresholds are the pass or fail criteria you define for your test metrics, and if the system does not meet one, the test finishes with a failed status. A threshold judges the run. An abort rule ends it early. A step-level threshold matters because an average over the whole flow hides one slow call behind many fast ones.

These apply to every type in the table above. A soak test and a spike test still need a rule that says what a pass looks like, or the result is a chart and an argument.

Running API performance tests on real flows

For an API, the unit of a good test is the flow, not the endpoint. A test that hits one URL in a loop measures a path no user takes, and it skips the step where state passes from one call to the next. Log in, take a token, read a list, create something, read it back: that chain is what real traffic looks like.

When a flow is slow under load, look at its steps. A flow that is slow overall almost always has one step that is slower than the rest, and that step is where the work is. The post on load testing multi-step flows covers the mechanics, including carrying tokens and ids from one call to the next.

Which test to run first

Most teams do not need every type on day one. A sensible order follows the questions in the table.

  1. Smoke. Run a few users first, so a broken script does not waste a long test.
  2. Load. Set the baseline at the traffic you expect.
  3. Stress or breakpoint. Once the baseline is clean, find where the system gives out and how it fails.
  4. Soak. Add it when you suspect slow degradation, such as a leak that only shows after hours.
  5. Spike. Add it when traffic arrives in surges, such as a launch or a campaign email.

Each step answers a question the one before it cannot. A load test that passes tells you nothing about a leak, and a soak test that passes tells you nothing about a surge. If you are unsure, stop after step two and let a real problem pick the next test.

Where Stresseur fits

In Stresseur any recorded flow can also run as a load test. Without a profile file you get Smoke, Load, Stress, Spike, Soak and Breakpoint. The load testing page shows a flow running under load, and the load test docs have the profile reference.

The same flows run in CI. The workflow runs each flow 13 times, 1 warm-up and 12 measured, and uploads per-step timings. Stresseur flags a flow when its median is at least 10% slower than the baseline and outside its normal run-to-run variation. The Stresseur check reports on each pull request, with one check and one comment updated in place on each push.

One limit: scheduled soak tests are not available. If you are comparing tools first, API load testing tools for small teams lines up the options.

Conclusion

Performance testing is the discipline, and load testing is one test inside it. Load testing asks how the system behaves at the traffic you expect. Stress, soak and spike change the shape of that traffic to find limits, slow leaks and sudden surges. Start with a load test, read p95, error rate and requests per second, and add the other types when a question needs them. Test flows, not single endpoints.

Frequently asked questions

Is load testing a type of performance testing?

Yes. Load testing is one area of performance testing. Within it, a load test assesses how the system performs under expected normal conditions.

What is the difference between load testing and stress testing?

A load test assesses performance under expected normal conditions. A stress test assesses performance at the system's limits, when load exceeds the expected average.

What is an example of a performance test?

A spike test is one example: it validates behavior and survival through sudden, short and massive increases in activity. A soak test is another: it assesses reliability over extended periods.

What are the main metrics in performance testing?

Response time is the time from sending a request to receiving a response. Throughput is the number of transactions or requests a system can handle per second. Error rate is the percentage of requests that fail or return errors.

Can I run performance tests on every pull request?

You can time your flows on every pull request. Stresseur's CI workflow runs each flow 13 times, 1 warm-up and 12 measured, and uploads per-step timings. Any recorded flow can also run as a load test.

All posts