Should I use average or p95 latency?

Use average for baseline and p95 for safer capacity assumptions.

Does this include queueing effects?

No, queueing and saturation dynamics should be validated in tests.

API throughput calculator

Q: What safety factor should I use?

70% to 80% is a practical baseline for planning.

Estimate theoretical and safe throughput for APIs using latency and effective concurrency parameters.

Calculator

Average latency (ms)WorkersConcurrent connections per workerSafety factor (%)

Results

Total concurrency

100

Theoretical max throughput

400 RPS

Safe throughput

320 RPS

Requests per minute (safe)

19,200

Real throughput depends on CPU saturation, queueing, GC, IO limits, retries and network conditions.

How throughput is estimated

We treat workers and connections as execution slots. Throughput is slots divided by latency in seconds.

The formula

total_concurrency = workers * connections_per_worker
throughput_max = total_concurrency / latency_seconds
throughput_safe = throughput_max * safety_factor

Use safe throughput for test planning and alert thresholds.

Safe throughput vs theoretical throughput

Theoretical values are upper bounds. Safe throughput leaves room for spikes, noisy neighbors and latency tail behavior.

FAQ

What safety factor should I use?

70% to 80% is a practical default for early planning.

Should I use p95 latency?

For conservative planning, yes. For average-case projections, use average latency.

Does this include queue wait time?

No, queue delay should be evaluated separately in load tests.