SLA Uptime Calculator

“Three nines” sounds close to perfect until you convert it to minutes. Service level agreements promise a percentage of time, and the number only becomes meaningful when it is translated into the longest outage you would have to accept. The page does that conversion in both directions and adds a model for combining components.

From percentage to time

The share of time a service may be down is 100 minus the availability. For 99.9 percent that is 0.1 percent. A day has 86,400 seconds, so the allowance is 86.4 seconds, which the page shows as 1 min 26.4 s. Scaling up gives 10 min 4.8 s a week, 43 min 12 s in a 30-day month and 8 h 45 min 36 s in a 365-day year. The same method for 99.99 percent gives 4 min 19.2 s per month and 52 min 33.6 s per year, and 99.999 percent leaves about 25.9 seconds and 5 min 15.4 s.

The four modes

“Downtime allowed by an uptime percentage” is the default. “Uptime percentage from downtime” works the other way: enter minutes of outage in a period and see the achieved availability. “Several components that all must work (in series)” multiplies availabilities, so a chain of a load balancer, an application server and a database each at 99.9 percent yields 99.7 percent, noticeably worse than each part. “Redundant components (in parallel)” combines them as 1 − (1 − a)(1 − b); presets include three nines, four nines and two servers in parallel.

Reading the contract, not just the number

The same percentage can mean very different protection. Contracts differ in the measuring period (month or year), in what counts as downtime (complete unavailability or degraded response), in exclusions for planned maintenance, and in the remedy, usually a service credit that is far smaller than the cost of an outage. A higher percentage over a shorter period is stricter: 99.9 percent per month allows 43 minutes at a time, while 99.9 percent per year allows nearly nine hours in total. Check the definitions before comparing offers.

A worked example

At 99.9 % the page lists 1 min 26.4 s per day, 10 min 4.8 s per week, 43 min 12 s per month, 2 h 11 min 24 s per quarter and 8 h 45 min 36 s per year. Choose “two servers in parallel” at 99 percent each, and the combined availability is 99.99 %, which allows only 4 min 19.2 s a month. The series example is the opposite: two components at 99.9 percent give 0.999 × 0.999 = 99.8001 percent.

Limits

The parallel model assumes independent failures. In real systems the two servers may share a power feed, a switch, a software bug or an operator, which brings the true figure well below the calculation. The page uses fixed month and year lengths, so it does not account for leap years or the exact number of days in a month. It converts and combines numbers; it does not measure a service. Uptime of a link also depends on the physical path, which the Fibre Loss Budget Calculator helps check.

Related tools

• RAID Calculator shows how many disk failures an array survives.
• Fibre Loss Budget Calculator checks an optical link has margin.
• DNS TTL Calculator plans record changes so migrations do not cause downtime.
• ISO 8601 Duration Calculator converts outage lengths into standard duration notation.