99.9% uptime:
how much downtime is that?
The allowance per day, week, month and year is below — type any other percentage to compare it, around the clock or business hours only.
The nines, translated into clock time.
Every nine you add divides the allowance by 10. The row you promise is the row you have to engineer for.
Allowed downtime per window
365-day year · month = year ÷ 12| SLA | Grade | day | week | month | year |
|---|---|---|---|---|---|
| 99% | two nines | 14m 24s | 1h 40m 48s | 7h 18m 0s | 3d 15h 36m |
| 99.5% | two nines | 7m 12s | 50m 24s | 3h 39m 0s | 1d 19h 48m |
| 99.9% | three nines | 1m 26s | 10m 4s | 43m 48s | 8h 45m 36s |
| 99.95% | three and a half nines | 43s | 5m 2s | 21m 54s | 4h 22m 48s |
| 99.99% | four nines | 8.6s | 1m 0s | 4m 22s | 52m 33s |
| 99.999% | five nines | 0.9s | 6s | 26s | 5m 15s |
How to calculate downtime — and uptime
Two directions, one formula. Forward, from the SLA to the allowance:
- Pick the window. A week is 604,800 s, a month 2,628,000 s, a year 31,536,000 s.
- Multiply by what’s left of 100%. At 99.9%, that’s 0.001 — so a week allows 604,800 × 0.001 = 604.8 s ≈ 10 minutes.
Backwards, from an outage to your actual uptime: subtract the downtime from the window, divide by the window. A 43m 48s outage in a month is (2,628,000 − 2,628) ÷ 2,628,000 = 99.90% — three nines, spent to the last second.
Work it backwards
downtime → uptime %Common uptime calculator questions
Per day: 1m 26s. Per week: 10m 4s. Per month: 43m 48s. Per year: 8h 45m 36s. (Using a 365-day year and month = year ÷ 12; tables using a 365.25-day year land a few seconds higher.) The intuition to keep: three nines allows one bad deploy a month — caught late.
Multiply the window by what’s left of 100%. For 99.9% over a week: 100% − 99.9% = 0.1% = 0.001, and 604,800 seconds × 0.001 = 604.8 seconds ≈ 10 minutes. That’s the entire formula — the calculator above just runs it across every window at once.
Backwards: uptime % = (window − downtime) ÷ window × 100. Suppose you had one 26-second blip today: (86,400 − 26) ÷ 86,400 = 99.97% — which already misses 99.99% for the day, because four nines allows only 8.6 seconds per day. This is why 99.99% is an engineering commitment, not a settings toggle.
The percentage is measured only against the hours you’re open — say Mon–Fri 9–18, which is 45 of the week’s 168 hours. Two consequences pull in opposite directions. The same 99.9% now allows only 11m 43s of daytime downtime a month. But nights and weekends vanish from the ledger: a 48-hour Saturday outage scores 100%. The two-clocks section of the result walks through one outage graded both ways.
Whatever the contract says — and that’s the point. Most SLAs exclude announced maintenance windows from the calculation, which is reasonable, but it makes two numbers worth checking: how much maintenance can be scheduled per month, and how much notice is required. An SLA with unlimited, same-day “planned maintenance” is a percentage with an escape hatch.
Treat both as flags. No serious SLA promises 100%: physics disagrees. When you see it, the definitions section is doing the real work — read what doesn’t count as downtime. And five nines is 26 seconds of downtime a month. That is telecom-switch territory, and it assumes automated failover: no human notices, opens a laptop and fixes anything in 26 seconds. Casual claims of either usually mean the exclusions are doing the heavy lifting.
The one your architecture and your on-call rotation can honor. Each nine you add cuts the allowance to a tenth, and the engineering to hold it climbs with it. For most SaaS, a 99.9% monthly you can keep beats an aspirational 99.99% you’ll breach: breaches cost credits, trust, and renewal conversations. And remember the chaining rule. Promise 99.9% on top of two dependencies that each promise 99.9%, and the math is already against you.
You can’t calculate your way to it — this page only tells you the allowance. The actual number comes from measuring: independent checks against your service, frequent enough that short outages can’t slip between them, logged so you can show the ledger when the SLA conversation happens. That’s what uptime monitoring is; Uptimia runs those checks from 171+ locations in 70+ countries, every 60 seconds on the Basic plan and every 30 from Professional up.
Keep exploring
99.9% allows 43m 48s a month.
Spent in one incident or a hundred small ones, it’s the same ledger. Per day that’s 1m 26s; per year, 8h 45m 36s. Below: the full ledger, the same percentage on the business-hours clock, and what one nine up or down would change.
Allowed downtime at 99.9%, window by window
Budgets are per window and sum across incidents — three short outages spend the same ledger as one long one.
The ledger — allowed downtime at 99.9%
24 × 7What 0.1% of a month looks like
drawn to scaleLooks harmless on a bar; feels different at 3 a.m. when it’s being spent all at once. SRE teams call this an error budget: something to spend deliberately on deploys and migrations — not to discover gone.
The formula, with your numbers
server uptime calculationSame three steps for any window — swap in the seconds that window holds.
Had an outage instead?
the reverse directionThis page starts from the promise. If you’re starting from the minutes you actually lost, the downtime calculator grades them, prices them in your numbers, and shows where they went.
Grade an outage insteadThe same 99.9%, on the 24 × 7 clock and the business-hours clock
Business hours (Mon–Fri 9–18) is 45 of the week’s 168 hours. Same percentage, 27% of the denominator — stricter when it matters, blind when it doesn’t.
Allowed downtime at 99.9%, both clocks
your active clock is highlighted| Window | 24 × 7 | Business hours |
|---|---|---|
| day | 1m 26s | 23s |
| week | 10m 4s | 2m 42s |
| month | 43m 48s | 11m 43s |
| year | 8h 45m 36s | 2h 20m 47s |
Two truths at once: the business-hours budget is far smaller during the day — and a weekend outage spends none of it. Neither clock is wrong; only one matches what your customers felt on Tuesday. Whichever you pick, name the clock in the contract — “99.9%” without a clock is an argument waiting to happen.
One outage, two report cards
The percentage didn’t change — the denominator did. Steal the wording: “99.9%, measured monthly over business hours (Mon–Fri 9–18 CET), planned maintenance excluded with 48 h notice, capped at 4 h/month.”
One nine up deletes 39m 25s of your 43m 48s monthly budget
The ladder is logarithmic and budgets are linear — a one-digit change in the promise is a 10× change in the engineering.
Your percentage, one step either way
÷ 10 per step| SLA | What changes | day | month | year |
|---|---|---|---|---|
| 99% | one step down — 10× the room | 14m 24s | 7h 18m 0s | 3d 15h 36m |
| 99.9% | your promise — three nines | 1m 26s | 43m 48s | 8h 45m 36s |
| 99.99% | one step up — 10× the discipline | 8.6s | 4m 22s | 52m 33s |
Promise the row your architecture and on-call rotation can honor — a kept 99.9% beats a breached 99.99%. And remember the chaining rule: two 99.9% dependencies already cap you at 99.8%.
What fits in 4m 22s a month
Four nines also assumes sub-minute detection: with 5-minute checks an outage can burn 300 seconds — 35 daily budgets — before the first failed probe.
Free tools are just the start.
Uptimia keeps your sites healthy.
Uptime, SSL, domain expiry, page speed, transactions — monitored from 171+ locations worldwide. Free for 30 days.