How it works
How the math works
Allowed downtime is the share of time the SLA leaves out: (100% − SLA) × the length of the period. At 99.9%, that is 0.1% of the time.
Periods are averages: 24 hours for a day, 7 days for a week, 30.44 days for a month, 91.31 days for a quarter and 365.25 days for a year, to account for leap years. Most calculators and contracts do the same.
The error budget
Allowed downtime can be seen as a budget: every outage spends part of it. Enter the downtime already seen this month to see what is left.
While there is budget left, a team can take risks: ship more often, migrate, experiment. Once it is spent, the team slows down and works on reliability. The practice was popularized by Google's SRE teams.
End-to-end availability
A service cannot be more available than what it depends on. When dependencies are in series, their availabilities multiply: a 99.9% service that depends on three 99.95% services is only up 99.75% of the time end to end.
To guarantee an SLA end to end, your service alone must do better than that SLA. The calculator shows what it needs, or warns you when it is impossible.