Availability is measured in nines. Here's the breakdown:
| Availability % | Downtime/Y | DownTime/Mo | Downtime/w | Downtime/d | Class |
|---|---|---|---|---|---|
| 55.5555555% ("nine fives") | 162.33 d | 13.53 d | 74.92 h | 10.67 h | |
| 90% ("one nine") | 36.53 d | 73.05 h | 16.80 h | 2.40 h | 1 |
| 95% ("one nine five") | 18.26 d | 36.53 h | 8.40 h | 1.20 h | |
| 97% | 10.96 d | 21.92 h | 5.04 h | 43.20 m | |
| 98% | 7.31 d | 14.61 h | 3.36 h | 28.80 m | |
| 99% ("two nines") | 3.65 d | 7.31 h | 1.68 h | 14.40 m | 2 |
| 99.5% ("two nines five") | 1.83 d | 3.65 h | 50.40 m | 7.20 m | |
| 99.8% | 17.53 h | 87.66 m | 20.16 m | 2.88 m | |
| 99.9% ("three nines") | 8.77 h | 43.83 m | 10.08 m | 1.44 m | 3 |
| 99.95% ("three nines five") | 4.38 h | 21.92 m | 5.04 m | 43.20 s | |
| 99.99% ("four nines") | 52.60 m | 4.38 m | 1.01 m | 8.64 s | 4 |
| 99.995% ("four nines five") | 26.30 m | 2.19 m | 30.24 s | 4.32 s | |
| 99.999% ("five nines") | 5.26 m | 26.30 s | 6.05 s | 864.00 ms | 5 |
| 99.9999% ("six nines") | 31.56 s | 2.63 s | 604.80 ms | 86.40 ms | 6 |
| 99.99999% ("seven nines") | 3.16 s | 262.98 ms | 60.48 ms | 8.64 ms | 7 |
| 99.999999% ("eight nines") | 315.58 ms | 26.30 ms | 6.05 ms | 864.00 us | 7 |
| 99.9999999% ("nine nines") | 31.56 ms | 2.63 ms | 604.80 us | 86.40 us | 8 |
💡 See classes
Every nine increases the cost of your service by a truly immeasurable amount, so be prepared for that as you discretely justify every nine specified in your SLAs and NFRs.
🗒️ See here and here for just a bit more on the cost of nines and justfying them.
Don't kill your devs/ops folks over a few superflous nines where it isn't necessary. Consider the following table that demonstrates varying availability targets by time and service type.
| Component Scope | 0600 - 1159 Local | 0000 - 0559 Local |
|---|---|---|
| UX Related Systems | 99.999% | 99.9% |
| Non-UX Related Systems | 99% | 99% |