Skip to content
Omnikalc

How many servers does your service need at peak?

Enter the peak users, requests per user, server capacity, target utilization and spares: get the number of servers, the load after a failure and the monthly cost.

Updated on
Your inputs

Value out of range

Value out of range

Value out of range

%

Value out of range

Value out of range

Value out of range

Results

Servers to provision

10

The fleet can lose one server and stay within a safe load
Peak load (requests per second)
2,000
Servers needed for the load at the target utilization
9
Utilization with all servers running
57.1%
Utilization after losing one server
63.5%
Monthly cost
$1,400

How many servers will your service need at peak? Too few and the site slows to a crawl; too many and money is wasted. This calculator works from the peak load and the capacity of one server to give the number of servers needed for a target utilization, adds spares for redundancy, and shows the load that remains if a server fails.

How the calculation works

  • Peak load (requests per second) = concurrent users × requests per user per minute ÷ 60
  • Servers needed = peak load ÷ (capacity × target utilization), rounded up
  • Servers to provision = servers needed + spares
  • Utilization after a failure = peak load ÷ ((servers − 1) × capacity)
  • Cost = servers × price per server

Example

20,000 concurrent users send 6 requests a minute each: a peak of 2,000 requests per second. One server handles 350 and should not exceed 65%, so 9 servers are needed, and 10 with one spare. With all servers running, utilization is 57.1%; after losing one, it is 63.5%. At $140 per server a month, the cost is $1,400.

Why it matters

Capacity planning avoids outages at peak times, sets the budget for hosting and shows how much redundancy is worth. The calculation also helps compare options such as fewer large servers against many small ones, since the loss of one server weighs differently on the remaining capacity.

Good to know

Real traffic is uneven, with peaks and bursts, and requests differ in cost, so the average per server is an estimate. Auto-scaling, caching and a content delivery network change the numbers. Always validate with a load test and monitoring before relying on the result. Capacity is also not the only limit: databases, network bandwidth and third-party services can saturate before the web servers do, so check the slowest component in the chain. Plan for growth by repeating the calculation with the traffic expected over the next year. This tool is for education.

The MTBF calculator gives availability from failure and repair times, the uptime SLA calculator converts a percentage into allowed downtime, and the Little’s law calculator relates work in progress to delivery time.

Frequently asked questions

How is the peak load estimated?

Concurrent users multiplied by requests per user per minute, divided by 60, gives requests per second. Use peak figures from monitoring rather than the average over a day, since peaks size the system.

Why not run servers at 100%?

Near saturation, queues build up and latency rises sharply. Leaving 30 to 40 percent headroom absorbs spikes, slow requests, deployments and growth.

What are spare servers for?

They keep the service up when a server fails or is being updated. With N servers needed, N + 1 is the usual minimum, and critical services often use N + 2 or spread servers over several zones.

How do I find the capacity of one server?

By load testing: increase the traffic until latency or errors exceed your limits, and take the request rate just below. Estimates from specifications are optimistic.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.