What the nines really allow
| Uptime | Per month | Per year |
|---|---|---|
| 99% | 7h 18m | 3 days 15h 39m |
| 99.5% | 3h 39m | 1 day 19h 49m |
| 99.9% ("three nines") | 43m 50s | 8h 45m 58s |
| 99.95% | 21m 55s | 4h 22m 59s |
| 99.99% ("four nines") | 4m 23s | 52m 36s |
| 99.999% ("five nines") | 26s | 5m 16s |
The difference between 99.9% and 99.99% looks tiny, but it is the difference between a 40-minute outage being within budget and a 5-minute one being a breach.
How it is calculated
Allowed downtime = (1 - uptime %) × length of the period
For 99.9% in a month: 0.001 × 30.44 days × 24 hours × 60 minutes = 43.8 minutes. The calculator uses an average month (365.25 / 12 days), as most SLA contracts do. Some contracts measure calendar months instead, which makes February slightly stricter.
To go the other way, enter the downtime you had in a month and the calculator gives the uptime you delivered.
Chained services
If your application needs a database, a payment gateway and a hosting provider, the uptime of the whole is the product of the parts:
99.9% × 99.95% = 99.85%
Each dependency in the chain lowers the total, so you cannot promise customers 99.99% on top of a provider that only offers 99.9%. Redundancy (two independent servers, either of which can serve) works the other way and raises availability.
Reading cloud SLAs
Cloud providers usually promise one figure per service and per setup: a single virtual machine might be covered at 99.5% or 99.9%, while machines spread across availability zones get 99.99%. The remedy for a breach is normally a service credit (a percentage off that month's bill), not compensation for your lost business. Planned maintenance is often excluded.
What a realistic target looks like
- A small business website on one VPS: 99.5-99.9% is realistic with monitoring and quick restarts.
- A SaaS product with a load balancer and two app servers: 99.9-99.95%.
- 99.99% and above needs several zones, automatic failover, careful deployments and someone on call. Each extra nine costs far more than the last.
Whatever you promise, measure it. An external uptime monitor that checks every minute gives you the real figure, and the history you need when a customer asks.