"99.9% uptime" reads like an A+ on a report card. In practice it means your system is allowed to be down for 8 hours and 46 minutes every year — and if you're the one negotiating (or signing) the SLA, that gap between the percentage and the actual downtime is where the real conversation happens.
The formula behind every "nine"
The math is one line: downtime = (1 − uptime%) × period length. At 99.9% uptime, the unavailable fraction is 0.1% (0.001). Multiply that by a 365.25-day year — 31,557,600 seconds — and you get roughly 31,558 seconds, or 8h 46m.
Flip it around and you can go the other way: if a contract says "no more than 5 minutes of downtime per month," divide 5 minutes by a 30-day month to get the fraction of time unavailable, subtract from 1, and you've derived the SLA percentage that promise actually implies.
Two length conventions that trip people up
Before the math means anything, you need to agree on what a "year" and a "month" are:
- Year = 365.25 days (a Julian year, accounting for leap years), so a quarter works out to 91.31 days — not a clean 91 or 92.
- Month = 30 days flat, week = 7 days, day = 24 hours — the standard flat assumptions used in most SLA contracts and SRE references like uptime.is.
If your contract defines the month or year differently (a fiscal quarter, a 28-day billing cycle), the percentage-to-downtime conversion doesn't change — you just scale the period length before multiplying.
The table everyone eventually memorizes
| SLA uptime | Per year | Per month | Per day | | --- | --- | --- | --- | | 99% (two nines) | 3d 15h 39m | 7h 12m | 14m 24s | | 99.5% | 1d 19h 49m | 3h 36m | 7m 12s | | 99.9% (three nines) | 8h 45m 58s | 43m 12s | 1m 26s | | 99.95% | 4h 22m 59s | 21m 36s | 43s | | 99.99% (four nines) | 52m 36s | 4m 19s | 9s | | 99.999% (five nines) | 5m 16s | 26s | 864ms |
Notice the pattern: each additional nine cuts the downtime budget by roughly 10x. That's not a coincidence — it's the same formula with one more zero after the decimal point — but it's exactly why the jump from 99.9% to 99.99% is where infrastructure costs start climbing fast. Four nines means your daily budget is measured in single-digit seconds, which rules out anything but automated failover; a human paged at 2am can't diagnose and fix an outage in 9 seconds.
Error budgets: the SLA's flip side
An error budget is just the downtime allowance reframed as something you spend. A 99.9% monthly SLA gives you 43m 12s to burn across deploys, incidents, and planned maintenance. Track how much you've used and you get a concrete, numeric answer to "is it safe to ship this risky change today" — if you're at 40 of your 43 minutes with a week left in the month, the answer is no, not until the budget resets.
This is also the honest way to push back on unrealistic SLA asks: "five nines" sounds like a marketing bullet point until someone points out it means 5 minutes and 16 seconds of total downtime for the entire year, including planned maintenance.
Do the math for your own numbers
Our Uptime & SLA Calculator runs all three directions — SLA percent → downtime budget, downtime target → required SLA percent, and error budget remaining — entirely in your browser. Type in your actual SLA tier, or the downtime you can tolerate, and skip re-deriving the formula by hand every time a contract negotiation brings it back up.