Virtual Users (VU) & RPS Load Test Calculator
Size your HTTP load tests accurately before execution. Compute required concurrent Virtual Users from target Requests Per Second (RPS), p50/p99 latency, and think time using Little’s Law.
λ (Arrival rate): How many HTTP requests per second your endpoint must process.
W (Service time): Typical HTTP round-trip latency under normal operating conditions.
Models how many Virtual Users get tied up if your backend slows down to p99 latency.
Z (Think time): Pause between sequential requests per Virtual User (0ms for raw API benchmarks).
Total execution window used to estimate cumulative request volume and plan fit.
Steady-State VUs
14
At p50 (150ms)
Recommended VUs
18
+25% headroom buffer
Starter Plan
For solo developers and small projects
- Up to 50 VUs / 50 RPS
- Up to 5 min per test run
In queueing theory, Little’s Law (L = λ × W) states that the long-term average number of items in a stationary queueing system (L) equals the long-term average effective arrival rate (λ) multiplied by the average time that an item spends in the system (W).
When a Virtual User pauses for client think time (Z) between requests, that VU slot remains occupied for W + Z seconds per request cycle. Ignoring response time or think time when picking VU counts leads to either severely under-driving your target API or exhausting worker sockets prematurely.
| Target RPS | 50 ms API | 150 ms API | 300 ms API | 1,000 ms API |
|---|---|---|---|---|
| 10 RPS | 1 VU | 2 VUs | 3 VUs | 10 VUs |
| 50 RPS | 3 VUs | 8 VUs | 15 VUs | 50 VUs |
| 200 RPS | 10 VUs | 30 VUs | 60 VUs | 200 VUs |
| 1,000 RPS | 50 VUs | 150 VUs | 300 VUs | 1,000 VUs |
Frequently Asked Load Sizing Questions
Using Little’s Law from queueing theory: VU = RPS × ((ResponseTime_ms + ThinkTime_ms) / 1000). For example, if your API target is 100 RPS, median response time is 200 ms, and think time is 0 ms, each Virtual User completes 5 requests per second (1000 / 200), requiring 20 concurrent VUs.
Steady-state VU math uses median (p50) latency. In real HTTP tests, tail requests (p95 and p99) take significantly longer, tying up active Virtual User threads. Adding a 25% VU buffer prevents your load generator from becoming the bottleneck when latency fluctuates.
Think time increases the total cycle duration (W + Z) of each Virtual User without generating additional server load during the pause. Adding 800 ms of think time to a 200 ms API call increases cycle time from 0.2s to 1.0s, requiring 5× more concurrent VUs to sustain the same RPS.