Uptime and SLA calculator: what the nines allow
How much downtime an SLA percentage allows per day, month and year; the uptime you actually delivered; and what a whole system adds up to when every part has its own SLA.
Downtime allowed by an uptime percentage
Type a percentage or pick one. A month is a twelfth of a 365-day year.
Uptime from the downtime you had
Add up the minutes the service was down, and pick the period they fell in.
Availability of a whole system
List every part a request needs, with its SLA and how many copies can each take the load on their own. Parts in a chain multiply; redundant copies cover for each other.
The nines at a glance
| Uptime | Name | Per day | Per week | Per month | Per quarter | Per year |
|---|---|---|---|---|---|---|
| 99% | Two nines | 14m 24s | 1h 40m | 7h 18m | 21h 54m | 3d 15h 36m |
| 99.5% | — | 7m 12s | 50m 24s | 3h 39m | 10h 57m | 1d 19h 48m |
| 99.9% | Three nines | 1m 26s | 10m 5s | 43m 48s | 2h 11m | 8h 45m |
| 99.95% | — | 43s | 5m 2s | 21m 54s | 1h 5m | 4h 22m |
| 99.99% | Four nines | 9s | 1m | 4m 23s | 13m 8s | 52m 34s |
| 99.999% | Five nines | 864 ms | 6s | 26s | 1m 19s | 5m 15s |
What the numbers don't tell you
An SLA is a refund policy, not a promise. When a cloud provider misses its SLA, you get service credits, a percentage of that month's bill for the affected service. Your lost sales and your customers' goodwill are not covered.
The provider's SLA is for their part only. AWS's region-level SLA for EC2 is 99.99%, but a single instance is covered at 99.5%. A site on one instance, one database and one deploy pipeline inherits the weakest of them, as the system calculator above shows.
Most downtime is self-inflicted. Failed deploys, expired certificates, full disks, a forgotten DNS change and patching without a plan cause more outages on small estates than hardware or provider failures do.
You can't report uptime you don't measure. Check from outside, every minute, from more than one place, and keep the record. It is what answers a customer's "what was your uptime last quarter?"
Getting to three or four nines
99.9% a month is reachable for most small SaaS products and agency platforms: two web servers behind a load balancer, a managed or replicated database, monitoring that wakes someone, and changes made in a planned window with a way back.
99.99% leaves about four minutes a month, which no human response time fits into. Every part needs automatic failover, including the database, deploys must be zero-downtime, and the failover has to be tested, not assumed.
Before promising either, write down what counts as downtime, how it is measured, and what is excluded. Those three lines matter more in a dispute than the percentage.
Questions
How much downtime is 99.9% uptime?
99.9% ("three nines") allows 8 hours 45 minutes of downtime a year, 43 minutes 48 seconds a month, or 1 minute 26 seconds a day.
How much downtime is 99.99% uptime?
99.99% ("four nines") allows 52 minutes 34 seconds a year, or 4 minutes 23 seconds a month. That is roughly one short restart a month, with no room for a failed deploy.
How do you calculate uptime percentage?
Uptime is the time the service was up, divided by the whole period, times 100. 30 minutes down in a 30-day month is (43,200 − 30) ÷ 43,200 = 99.93%.
How do I work out the availability of a whole system?
Multiply the availability of every part a request needs: three parts at 99.9% each give 99.7% together. Redundant copies work the other way: two servers at 99.5% either of which can serve gives 1 − (0.005 × 0.005) = 99.9975%.
Does planned maintenance count as downtime?
It depends on the SLA. Many contracts exclude announced maintenance windows, which is why the measured uptime and the contractual figure can differ. Your users count it either way.
What uptime should a small SaaS promise?
99.9% a month is the usual starting point, and it is achievable on a single well-run region with a redundant web tier. Promising 99.99% needs automatic failover for every part, including the database, and someone answering alerts at night.
Want to know what your setup really delivers?
The £250 GUARD review checks the parts that cause most outages, from certificates and backups to patching and alerts, and tells you what to fix first.