Uptime calculator

What each nine buys you in minutes

Percentages flatter. 99% looks like a top mark and allows 7.2 hours of downtime in a 30 day month, or 3.65 days a year. Move to 99.9% and the monthly allowance becomes 43.2 minutes, which is one unlucky deployment. 99.95% is 21.6 minutes, 99.99% is 4.3 minutes and 99.999% is about 26 seconds a month, or 5.3 minutes a year. Each extra nine divides the budget by ten, and the engineering cost does not grow in a straight line.

The calculator uses a 30 day month and a 365 day year. A calendar month of 31 days gives 44.6 minutes at 99.9%, February gives 40.3, so a contract that defines the month differently will produce slightly different figures. Check how your SLA counts before comparing.

Stacking components lowers the number

A site is only as available as the things it depends on in series. If your host promises 99.9%, your DNS provider 99.99% and your payment gateway 99.95%, a checkout that needs all three works about 99.84% of the time (0.999 x 0.9999 x 0.9995), which is roughly 69 minutes a month. Two components at 99.9% each give 99.8%, double the downtime of either alone.

Redundancy reverses the arithmetic, but only when failures are independent. Two servers each at 99.9% behind a load balancer give 99.9999% on paper (the chance that both are down at once is 0.001 x 0.001), but the load balancer, the shared database and the deployment process remain single points of failure, and that is where the real number ends up.

Choosing what to promise a client

Promise what you control. A shared hosting account rarely gets better than 99.9% in a good month, so writing 99.99% into a care plan is a refund waiting to happen. A safer structure is a response commitment (acknowledged within 30 minutes during business hours), a reporting commitment (a monthly uptime figure) and a target rather than a guarantee. Then show the measured number: if the site delivered 99.94% last month, say that and list the two incidents behind it.

Exclusions matter as much as the headline. Planned maintenance, upstream outages at a third party the client chose, and DDoS events are commonly carved out, and each carve-out should be written down in the same document as the percentage.

Measured uptime depends on how you measure

Checking every 5 minutes cannot see a 90 second outage at all, and one failed check is conventionally counted as the whole interval. At a 5 minute interval a single failure costs 0.012% of a 30 day month; at a 1 minute interval the same incident costs 0.002%. Higher frequency does not change what happened, it changes how precisely it is recorded.

In Tenium an outage begins when the confirming regions agree and ends when they agree on recovery, and maintenance windows are left out, so the figure shown on a status page and in a report is derived from the incident list rather than from raw failed checks. Status pages show it per monitor, and the Free plan includes 1 of them.

Using the page for planning

Enter the target from a contract to see the budget, then divide it by your typical incident length. A 43 minute monthly allowance holds one 30 minute incident but not two. If your last three outages lasted 25, 40 and 70 minutes, no setting of the dial will make 99.9% realistic until recovery time shrinks. The number to improve is usually the time to notice, which is why check interval and alert routing matter more than another layer of hardware.