Uptime calculator
Turn an uptime percentage such as 99.9% or 99.99% into the down time it allows a year, a month, a week and a day, and see what running two or three independent copies in separate zones or regions does to it, in theory.
- Down time allowed a year at 99.9%
- 8 hr 45 min 36 sec
- That’s 43 min 48 sec a month or 1 min 26 sec a day. Two copies that never fail together: 32 sec a year, in theory.
| Down time | One copy | Two copies, in theory |
|---|---|---|
| Uptime | 99.9% | 99.9999% |
| A year | 8 hr 45 min 36 sec | 32 sec |
| A month | 43 min 48 sec | 2.6 sec |
| A week | 10 min 5 sec | 0.6 sec |
| A day | 1 min 26 sec | 86 ms |
A year here is 365 days, a month is a twelfth of that (30 days 10 hours) and a week is 7 days. Down time is to the nearest second, or the nearest tenth of a second under 10 seconds. An uptime with more than eight decimal places is shown as a count of nines: 99.9999999999% is 12 nines.
The copies column is a theoretical figure: 1 − (1 − uptime)copies. It only holds if the copies fail independently, never for the same reason at the same time, and if switching from one to another is instant. Real outages often break that: a bad update pushed to every region at once, or a fault in something they all share, takes every copy down together. The Google figures in the notes are the monthly uptimes in its Compute Engine and Spanner service level agreements, checked on 20 October 2026. Missing one earns customers a credit on the bill; it isn't a guarantee the service stays up.
What goes in
Type an uptime percentage, or pick one of the common figures: 99%, 99.9%, 99.95%, 99.99% or 99.999%. People call these two nines, three nines and so on. Then choose whether to compare it with two or three copies of the same service, each in its own zone or region.
What the result means
The headline is how much down time a year that uptime allows. 99.9% sounds close to perfect, but it’s 8 hr 45 min 36 sec a year, or 43 min 48 sec a month. 99.99% is 52 min 34 sec a year, and 99.999% is 5 min 15 sec.
The table also works out the same for the copies. Two copies at 99.9% that never fail at the same time are both down only 0.1% × 0.1% of the time, so together they’re up 99.9999%: 32 sec a year. The sum is 1 − (1 − uptime)copies.
That copies figure is theoretical. It assumes each copy can run everything on its own, the switch between them is instant, and they never fail for the same reason at the same time. Real outages often break the last one: a bad update pushed to every region at once, or a fault in something every copy relies on, takes them all down together. So treat it as the best case.
Google’s own figures
Google Cloud’s service level agreements, checked on 20 October 2026, promise on its Premium network tier a monthly uptime of 99.9% for a single Compute Engine virtual machine, 99.95% for a single memory-optimised one and 99.99% for virtual machines “placed across two or more Zones in the same Region”. Its Spanner database promises 99.99% in one region and 99.999% across more than one. If Google misses one, customers get a credit on their bill. It isn’t a promise the service never goes down for longer.
How it’s worked out
A year here is 365 days, a month is a twelfth of that (30 days 10 hours) and a week is 7 days. Down time is the period × (1 − uptime), shown to the nearest second, or to a tenth of a second under 10 seconds. Google’s agreements measure uptime over each calendar month, so a 31-day month allows a little more than this and February a little less.
Sources
- Google Cloud: Compute Engine Service Level Agreement, checked 20 October 2026
- Google Cloud: Spanner Service Level Agreement, checked 20 October 2026
- Google Cloud: Regions and zones