SLA uptime calculator

Uptime percentages sound interchangeable until they are minutes. Enter a percentage to see the downtime it permits per day, week, 30-day month, quarter and year.

The reference table

Downtime allowed at each level, rounded to the second. A month is taken as 30 days and a year as 365, which is what most contracts mean unless they say otherwise.

What the nines are worth in practice

The jump from 99% to 99.9% is the one that changes how a service is run. At 99% a person can notice an outage, open a laptop and fix it inside the budget. At 99.9% the monthly allowance is 43 minutes, which is roughly one unplanned reboot and the checks afterwards, so the response has to be automatic rather than human. At 99.99% the allowance is four minutes a month: nobody is being paged and fixing it in time, so it has to survive the failure without intervention, which means redundancy rather than monitoring.

That is why the price of each extra nine rises so sharply. You are not buying a better-maintained server, you are buying a second one.

Stacked dependencies multiply, they do not average

This is the arithmetic most SLAs are read without. When a service needs several components to be up at once, the availabilities multiply. Three components at 99.9% each give 0.999 x 0.999 x 0.999, which is 99.7%: about 2 hours 10 minutes a month, not 43 minutes.

Count every link in the chain the client actually depends on: the application, the database, the hosting platform, DNS, the network path and any third-party API in the request. A stack of five components each promising 99.9% cannot honestly be sold to the client as 99.9%. If you need the number to come back up, the answer is redundancy on the weakest component, because removing a single point of failure changes the maths from multiplication to something much kinder.

Reading an SLA with this open

Check the measurement window first. Per month resets the meter monthly; per year lets one bad day eat the whole budget and still technically pass. Check what counts as down: scheduled maintenance windows usually do not count, and neither does anything the provider classifies as a force majeure or a customer-caused fault, so the measured figure is always kinder than the experienced one. Check whether it is measured from inside the provider's network or from where the user sits.

Then check the remedy. Service credits are typically a small percentage of one month's fee, not your losses, and on most contracts you have to claim them in writing within a set number of days or they are forfeited. Never promise a client tighter uptime than your own suppliers promise you.