Server Sizing Calculator

Target CPU Utilization % is the denominator the calculator divides your raw CPU need by, so a lower target — meaning you want more idle headroom — produces a larger recommendedVCPUs figure for the same traffic. Setting it to 50% instead of 70%, for example, provisions roughly 40% more CPU for identical concurrent-request load, trading cost for a bigger buffer against traffic spikes.

The instancesNeeded output is calculated from recommendedVCPUs divided by a typical 8-vCPU instance size, but the formula always returns at least 2 regardless of how small that division comes out — this models the standard high-availability practice of never running a production service on a single instance, since one node has no failover if it goes down.

The main recommendedVCPUs is sized for your entered concurrent users and target utilization as-is. Peak vCPUs reruns the same CPU-per-request math against double your calculated requests-per-second (peakRps = totalRps × 2) to show what capacity a traffic surge would demand — it's a separate 'what if' figure, not included in the main recommendation, so you'd need to provision toward it explicitly if you want standing headroom for a 2x spike.

It works backward through a form of Little's Law: concurrent users and their requests-per-minute rate produce requests per second, which combined with your average response time gives the number of requests in flight at any instant (concurrency = throughput × latency). Each of those in-flight requests is assumed to cost about 0.15 vCPU, a fixed compute-intensity figure meant for typical web/API work rather than CPU-heavy processing like video encoding or ML inference.