From percentage to minutes
The Google SRE book defines availability as “the fraction of the time that a service is usable” and notes that the industry speaks of “nines”: 99 percent is “2 nines” and 99.999 percent is “5 nines” (Service Level Objectives). Allowed downtime is (1 - availability) x period. A 30-day month has 43,200 minutes, so 99 percent allows 432 minutes, 99.9 percent allows 43.2 and 99.99 percent allows about 4.3.
Combining services
Services that must all work multiply: three at 99.9 percent give about 99.70 percent. Redundant services that are independent combine as 1 - (1 - A1) x (1 - A2), though shared infrastructure breaks the independence assumption.
Honest measurement
A period without a valid measurement is unknown, and counting it as up inflates the figure. A reported figure should come with its period, method and the minutes of unknown time. Google’s SRE book (Embracing Risk) describes both calculations: one based on the proportion of system uptime, and one on the request success rate, successful requests over total requests, which suits large services that are rarely fully down. The same chapter works one example: a system with a 99.99 percent target can be down for up to 52.56 minutes in a year.
A worked example
A client site has two outages in a 30-day month of 43,200 minutes: one of 18 minutes and one of 31. Monitoring also had no valid measurement for 10 minutes. Downtime is 49 minutes, so availability is (43,200 - 49) / 43,200, which is 99.89 percent. The 10 minutes are reported as unknown instead of being added to the uptime. A target of 99.9 percent allows 43.2 minutes, so this month missed it, and the report can say by how much and show what was and was not measured.
What to check
The definition of success, the measurement interval and how gaps are treated.
Sources
Read on . Standards and vendor pages change; confirm the current text before you rely on it.