The nines table
Allowed downtime for common SLA levels, 24/7, with an average month of 30.44 days and a year of 365.2425 days. Linked levels have their own page, and the nines of availability explains what each one takes.
| Availability | Per day | Per week | Per month | Per quarter | Per year |
|---|---|---|---|---|---|
| 90% | 2h 24m | 16h 48m | 3d 1h 2m | 9d 3h 8m | 36d 12h 34m |
| 95% | 1h 12m | 8h 24m | 1d 12h 31m | 4d 13h 34m | 18d 6h 17m |
| 99% | 14m 24s | 1h 40m 48s | 7h 18m 17s | 21h 54m 52s | 3d 15h 39m |
| 99.5% | 7m 12s | 50m 24s | 3h 39m 9s | 10h 57m 26s | 1d 19h 49m |
| 99.9% | 1m 26s | 10m 5s | 43m 50s | 2h 11m 29s | 8h 45m 57s |
| 99.95% | 43s | 5m 2s | 21m 55s | 1h 5m 45s | 4h 22m 58s |
| 99.99% | 8.6s | 1m | 4m 23s | 13m 9s | 52m 36s |
| 99.995% | 4.3s | 30s | 2m 11s | 6m 34s | 26m 18s |
| 99.999% | 0.9s | 6s | 26s | 1m 19s | 5m 16s |
| 99.9999% | 0.1s | 0.6s | 2.6s | 7.9s | 32s |
How the numbers are worked out
Allowed downtime is the part of the period the SLA doesn't promise: downtime = period × (1 − SLA). For 99.9% over a 30-day month that's 30 × 24 × 60 × 0.001 = 43.2 minutes, or 43m 12s.
Going the other way, uptime = (1 − downtime ÷ measured time) × 100. Excluded maintenance comes off the measured time before you divide. If the SLA covers business hours only, the measured time is just those hours, so the same percentage allows much less downtime.
Calculators disagree because they define a month differently: 30 days, the real calendar month (28 to 31 days), or an average month of 30.44 days. 99.9% allows 40m 19s in February and 44m 38s in a 31-day month. Whichever you use, write it into the agreement.
What counts as downtime?
Whatever the agreement says. Settle these before anyone has to argue about them:
- Partial outages. Is the site down if checkout fails but the home page loads?
- Slow responses. Many SLAs count a response slower than a set limit as down.
- Scheduled maintenance. Usually excluded if announced in advance, often up to a monthly limit.
- Regional failures. One network path failing isn't the same as the site being down. 12happy confirms every failure from a second location before it alerts, on every plan, so a blip on one route doesn't count as an outage.
Your check interval decides what you can prove
A monitor only sees an outage if a check runs during it. So the downtime you measure is only accurate to about one check interval, and an outage shorter than the interval can be missed entirely. We count a monthly budget as verifiable when it spans at least three intervals, borderline from one to three, and not verifiable below that.
| Checks | 99.9% (43m 50s/mo) | 99.95% (21m 55s/mo) | 99.99% (4m 23s/mo) | 99.999% (26s/mo) |
|---|---|---|---|---|
| Every 5 minutes · Free | Yes | Yes | No | No |
| Every minute | Yes | Yes | Yes | No |
| Every 30 seconds · Pro | Yes | Yes | Yes | No |
| Every 15 seconds · Agency | Yes | Yes | Yes | Borderline |
12happy checks every 5 minutes on Free, every 30 seconds on Pro and every 15 seconds on Agency. Compare plans
Composite SLAs: why your app can't beat its dependencies
If your site needs its hosting, its database and its DNS all to be up, its availability is at most the product of theirs. A 99.9% app on 99.95% hosting with 99.9% DNS can reach about 99.75%, not 99.9%. That's roughly 1h 50m of downtime a month instead of 44 minutes.
Redundancy works the other way. Two independent copies of a 99% component are both down only 0.01% of the time, so together they reach 99.99%. The catch is the word independent: two servers in the same data centre often fail together.
Monitor each dependency on its own as well as the site. When something breaks you'll know which part, and a status page can tell your users the same.
SLA, SLO, SLI and error budgets
An SLI (indicator) is what you measure, such as the share of successful checks. An SLO (objective) is the target you set yourself, such as 99.95%. An SLA (agreement) is the promise you make to a customer, usually looser than the SLO, with consequences if you miss it. The error budget is the downtime the SLO leaves you: spend it on releases and maintenance, and slow down when it runs out. Google's SRE book explains the idea in Embracing risk.
Putting uptime into a client contract
For agencies and freelancers who host or look after client sites, a clause usually covers five things:
- The target, such as 99.9% a calendar month.
- How it's measured: an external monitor, the check interval, and confirmation from a second location.
- Exclusions: maintenance announced 48 hours ahead, outages caused by the client or their other providers.
- Remedies: credit tiers, a cap, and a deadline for claiming.
- Reporting: a status page the client can check, and a monthly uptime figure.
Example: "The Provider will make the Website available 99.9% of each calendar month, measured by external checks at least every minute from two locations, excluding maintenance announced at least 48 hours in advance. If availability falls below 99.9%, the Client is entitled to a credit of 10% of that month's fee, or 25% below 99%, up to 50% of the fee, if claimed within 30 days." This is an illustration, not legal advice.
SLA credits in practice
Credits are usually tiers of the monthly fee, capped, and only paid if the customer claims them in time. They rarely cover what an outage really costs, so treat them as a signal of how seriously a provider takes uptime, not as insurance. To put a number on the real cost, use the downtime cost calculator.
Questions
How much downtime is 99.9% uptime?
About 43m 50s a month, 10m 5s a week, or 8h 45m 57s a year, using an average month of 30.44 days. On a 30-day month it is exactly 43m 12s.
What is the difference between 99.9% and 99.99% uptime?
Ten times less downtime: 99.9% allows about 43m 50s a month, and 99.99% allows about 4m 23s. Each extra nine divides the allowed downtime by ten.
How do I calculate uptime percentage from downtime?
Uptime % = (1 − downtime ÷ measured time) × 100. For example, 30 minutes of downtime in a 30-day month (43,200 minutes) is (1 − 30 ÷ 43,200) × 100 = 99.93%. Take any excluded maintenance out of the measured time first.
Does scheduled maintenance count as downtime?
Only if the agreement says so. Most SLAs exclude maintenance announced in advance, often within a window and up to a monthly limit. The calculator lets you exclude it, which removes it from the measured time.
What is a composite SLA?
The availability of a service that depends on several others. If all of them must be up, multiply their availabilities: 99.9% × 99.95% × 99.9% ≈ 99.75%, which is lower than any single one. Running two independent copies of a component raises its availability to 1 − (1 − A)².
How often do I need to check my site to prove an SLA?
A check only sees an outage if it runs during it, so your measurement is only accurate to about one check interval. To verify 99.99% (about 4m 23s a month) you need checks every minute or faster. Checks every 5 minutes can't tell 99.99% from 99.9%.
What is five nines?
99.999% availability: about 26s of downtime a month, or 5m 16s a year. It's a target for telecoms and core infrastructure. For most websites it costs far more than it earns.
Why do SLA calculators give different numbers?
They use different month and year lengths: a 30-day month, an average month of 30.44 days, or the real calendar month (28 to 31 days), and years of 365, 365.25 or 365.2425 days. The differences are small but real, so an SLA should say which it uses.
Know your real uptime, not just the target
12happy checks your sites around the clock, confirms failures from a second location, and alerts you by email, Slack, Telegram, push, SMS or webhook. Status pages are included on every plan.