Requests per second to concurrent users.

Worked example

Matches the AWS Lambda guide: 100 requests per second at a 500 ms average duration is 100 x 0.5 = 50 concurrent requests. Double the rate to 200 per second at 250 ms and concurrency is still 50. For users, add a 4.5-second think time: 100 x (0.5 + 4.5) = 500 concurrent users. With 20% headroom, plan for 60 requests in flight. In reverse, 500 users with a 5-second cycle (0.5 s response + 4.5 s think) generate 500 / 5 = 100 requests per second.

How it works

Requests in flight is Little's law: L = lambda x W with lambda in requests per second and W the average response time in seconds. For interactive users the law is N = X x (R + Z), where X is throughput, R response time and Z think time (the interactive response time law). It assumes a steady load and a closed set of users who each wait for a response before the next request.

Use the peak, not the day's average, and note that "concurrent users" depends on how you define a user session. If a session lasts minutes between page views, use a long think time. A load test with no think time creates far more load than the same number of real users.

FAQ

How do I convert requests per second to concurrent users?

Multiply requests per second by the time one user spends per request cycle: response time plus think time. 100 requests per second with a 5-second cycle is 500 concurrent users.

How many concurrent requests is 1,000 requests per second?

It depends on latency. At 200 ms average, 1,000 x 0.2 = 200 requests in flight. At 2 seconds it is 2,000.

What is think time?

The pause between a user receiving a response and sending the next request. Setting it to 0 models a client that fires continuously.

Why does my real concurrency not match the formula?

The formula uses averages over a steady period. Bursts, slow outliers and retries raise peak concurrency, which is why a headroom percentage is offered.