99.9% uptime in every period
The allowance is 0.1% of the period, so it grows and shrinks with the length of the month:
| Period | Length | Allowed downtime |
|---|---|---|
| Day | 24 hours | 1m 26.4s |
| Week | 7 days | 10m 4.8s |
| 28-day month | 672 hours | 40m 19.2s |
| 29-day month | 696 hours | 41m 45.6s |
| 30-day month | 720 hours | 43m 12s |
| 31-day month | 744 hours | 44m 38.4s |
| Average month | 30.42 days | 43m 48s |
| Quarter | 91.25 days | 2h 11m 24s |
| Year | 365 days | 8h 45m 36s |
| Leap year | 366 days | 8h 47m 2s |
A contract that measures each calendar month gives you the row for that month. February, with 28 days or 29 in a leap year, is the strictest month of the year. A provider that quotes 43 minutes a month is using the average month. For any other target, use the uptime calculator.
What 99.9% SLAs promise
Google Workspace commits to at least 99.9% uptime in every calendar month. Below that, customers get 3 days of service added as a credit, rising to 7 days below 99% and 15 days below 95%. Google counts a period as downtime when the user error rate of the web interface is above 5%.
Atlassian promises 99.9% a month on its Premium cloud plans and 99.95% on Enterprise, measured per calendar month in UTC. A minute counts as down when more than 5% of the requests in it fail. Atlassian's own example puts 60 down minutes in a 30-day month at about 99.86%, which earns a Premium customer a 10% credit.
Both define downtime by error rate. Up to 5% of requests can fail without that time counting against the SLA, so users can see errors in a month that still meets 99.9%.
99.9% vs 99.99% uptime
| Uptime | Per day | Per 30-day month | Per year |
|---|---|---|---|
| 99.5% | 7m 12s | 3h 36m | 1d 19h 48m |
| 99.9% | 1m 26.4s | 43m 12s | 8h 45m 36s |
| 99.95% | 43.2s | 21m 36s | 4h 22m 48s |
| 99.99% | 8.6s | 4m 19.2s | 52m 33.6s |
| 99.999% | 864ms | 25.9s | 5m 15.4s |
Each extra nine cuts the allowance by ten. In an average month, 99.9% leaves 43m 48s and 99.99% leaves 4m 22.8s. Eight 5-minute outages fit inside a 99.9% month, while a single 5-minute outage breaks a 99.99% one.
Amazon's EC2 SLA shows both ends in one contract: 99.5% for a single instance, and 99.99% only when your instances run across two or more availability zones in a region.
How far 43 minutes goes
One 45-minute outage breaks a 99.9% month of any length, because even a 31-day month only allows 44m 38.4s. Two 20-minute incidents fit in every month, February included.
Detection comes out of the same budget. With checks every 5 minutes, a failure can run for up to 5 minutes before the first failed check, which is 11% of an average month's allowance. Checks every minute cut that to 2%. Measuring against the target is covered in the nines of availability guide.
Sources
- Google Workspace SLA (99.9% per calendar month, credits of 3, 7 and 15 days, 5% error rate)
- Atlassian Cloud SLA (Premium 99.9%, Enterprise 99.95%, downtime minutes, credit tiers)
- Amazon EC2 SLA (99.99% region level, 99.5% instance level)