practice

Workload Model

A description of the traffic mix, arrival pattern and data distribution a load test reproduces, which determines whether the test's results mean anything.

A load test is only as good as its workload model. Most load tests produce reassuring numbers because they test something easier than production.

What a realistic model specifies:

Traffic mix — the proportion of each operation, taken from production telemetry rather than estimated. Reads and writes have entirely different characteristics, and a test that is 95% reads against a 70/30 production reality proves nothing about the write path.

Arrival pattern — real traffic is bursty. A constant arrival rate at the same mean produces shorter queues and better latency than reality.

Data distribution — this is where synthetic tests fail most often. Requests distributed uniformly across keys give an unrealistically high cache hit rate and no hot partitions. Production has a power law: a small number of items receive most of the traffic.

Data volume — testing against a small dataset gives query plans and index behaviour that do not survive production scale.

Think time and session behaviour — including connection reuse, which changes handshake costs dramatically.

Cold state — the first minutes after a deployment, with empty caches and unwarmed pools, are the worst case and are usually excluded from the measurement.