Reel · Chaos & Reliability · Oct 30, 2026
99.9% uptime = 8 hours 46 minutes of DOWN, every year
In short
99.9% uptime sounds perfect, but it allows 8h 46m of downtime a year: 43m 48s a month, about 86 seconds a day. 99.99% allows only 52m 34s a year, and every extra nine means 10x less downtime and a lot more cost. The real example: on Oct 19–20, 2025, a DNS race condition left DynamoDB's regional endpoint in AWS us-east-1 with an empty DNS record, from 11:48 PM to 2:20 PM PDT, about 14.5 hours. That's more than a whole year of 99.9% budget in one night (source: AWS's own post-event summary). Uptime isn't a promise. It's a budget.
Label from the original post: Voice and avatar are AI-generated.
Checklist: what to do about it
- Write your availability target down per service, and convert it to minutes per month.
- List your hard dependencies and their targets. You can't beat them without designing around them.
- Track the error budget you've spent this month, not just whether you're up right now.
- Agree with product what happens when the budget is spent: slow down and fix reliability.
- Before you promise another nine, price it: redundancy, regions, on-call and testing.
- Spend the budget on purpose (deploys, migrations, game days) instead of on surprises.
The NINES cheat sheet
Math: (1 − uptime) × period. Year = 365 days, month = 1/12 year.
| Uptime | Per year | Per month | Per week | Per day |
|---|---|---|---|---|
| 99% | 3d 15h 36m | 7h 18m | 1h 41m | 14m 24s |
| 99.5% | 1d 19h 48m | 3h 39m | 50m 24s | 7m 12s |
| 99.9% | 8h 46m | 43m 48s | 10m 4.8s | 1m 26.4s |
| 99.95% | 4h 23m | 21m 54s | 5m 2.4s | 43.2s |
| 99.99% | 52m 33.6s | 4m 22.8s | 1m 0.5s | 8.6s |
| 99.999% | 5m 15.4s | 26.3s | 6s | 0.9s |