99.9% uptime means a service may be unavailable for 0.1% of the time. That is 1 minute 26 seconds a day, 43 minutes 49 seconds in an average month and 8 hours 45 minutes 57 seconds a year. Each extra nine divides the allowed downtime by ten: 99.99% leaves 4 minutes 22 seconds a month, 99.999% only 26 seconds.
The percentage is the easy part. What decides whether a provider meets it is the small print: over which period it is measured, what counts as down, what is excluded and who does the measuring. This guide covers the numbers first, then the small print, then how to pick a target you can keep.
To work out any other target or period, use the SLA and uptime calculator. It also works the other way: enter the downtime you had and it tells you which SLA you stayed within.
Downtime allowed by each level of nines
The table uses the same math as the calculator: a year is 365.25 days (the average including leap years), a month is a twelfth of that (30.4375 days), and the allowed downtime is rounded down to the second, because rounding up would let you exceed the target.
| Uptime | Per day | Per week | Per month | Per year |
|---|---|---|---|---|
| 99% | 14m 24s | 1h 40m 48s | 7h 18m 18s | 3d 15h 39m 36s |
| 99.5% | 7m 12s | 50m 24s | 3h 39m 9s | 1d 19h 49m 48s |
| 99.9% | 1m 26s | 10m 4s | 43m 49s | 8h 45m 57s |
| 99.95% | 43s | 5m 2s | 21m 54s | 4h 22m 58s |
| 99.99% | 8s | 1m | 4m 22s | 52m 35s |
| 99.999% | under 1s | 6s | 26s | 5m 15s |
Some quick links for the common targets: 99.5%, 99.95% and 99.99%, each with the downtime per day, week, month, quarter and year.
Why your month may give a different number
Many contracts measure per calendar month, and calendar months are not the same length. At 99.9%, February (28 days) allows 40 minutes 19 seconds, a 30-day month 43 minutes 12 seconds and a 31-day month 44 minutes 38 seconds. The average month sits in between at 43 minutes 49 seconds. If you report against a contract, use the real length of the month. If you plan or compare targets, the average is the fairer number.
What the nines mean in practice
The step from one level to the next looks small on paper. In operations, each one changes how you have to work.
- 99% (7 hours a month): room for a bad evening or an unplanned restart that someone handles the next morning. Fine for internal tools and side projects, too loose for anything customers pay for.
- 99.5% (3.6 hours a month): one serious incident a month fits, if someone responds within the hour.
- 99.9% (44 minutes a month): the usual target for business websites and SaaS products. It needs alerting that reaches a person quickly and a way to roll back a bad deploy in minutes.
- 99.95% (22 minutes a month): one incident handled well, or two short ones. Manual response at night starts to be the limiting factor.
- 99.99% (4 minutes a month): shorter than it takes most people to wake up, open a laptop and read an alert. It takes redundancy and automatic failover, not faster humans.
- 99.999% (26 seconds a month): shorter than the gap between two checks for most monitoring. Few services achieve it, and fewer can prove it.
That last point matters for any target: your measurement has to be finer than the budget. With checks every five minutes, a single missed short outage can hide most of a 99.99% month. With checks every minute, you can measure 99.9% and 99.95% with confidence and 99.99% with care.
How SLAs measure uptime
Two providers can both promise 99.9% and mean very different things. When you read or write an SLA, look for these six points.
The measurement period
Most SLAs measure per calendar month, some per quarter or year. A longer period is kinder to the provider: 8 hours 45 minutes a year can be spent in one bad night and the yearly SLA is still met, while the monthly SLA for that month is broken many times over.
What counts as down
An error page and no answer at all are obvious. Less obvious: a page that loads in 30 seconds, a checkout that fails while the homepage works, or an API that answers 200 with an empty body. A good SLA names the components it covers and says whether slow responses and partial outages count.
Exclusions
The usual ones are scheduled maintenance announced in advance, problems caused by the customer, and events outside anyone’s control. Maintenance is the one to read closely. Without a notice period and a monthly cap, any outage can be called maintenance afterwards.
Where it is measured
A provider measuring from inside its own network will miss DNS failures, expired certificates and network problems that stop customers from reaching it. External checks see what customers see. The SLA should also say whether an outage seen from one location counts, or only one seen from several.
Minimum outage length
Many SLAs only count outages that last longer than a minute or are confirmed by a second check. That keeps a single lost packet out of the numbers, which is fair, as long as the threshold is short. A rule that only counts outages over 15 minutes can make a 99.9% promise nearly meaningless.
Who measures, and how you claim
Usually the provider’s own monitoring is the reference, and the customer has to request a credit within a set time, often 30 days after the month ends. Keeping your own external record of the service’s uptime gives you something to compare against.
Service credits
The usual consequence of a missed SLA is a service credit: a discount on a future invoice, scaled by how far below the target the month ended. An example of how a tiered scheme is often built:
| Monthly uptime | Credit (example) |
|---|---|
| 99.9% or more | None |
| 99.0% to 99.9% | 10% of the monthly fee |
| 95.0% to 99.0% | 25% of the monthly fee |
| Below 95.0% | 50% of the monthly fee |
Three things are typical. Credits are capped, often at the monthly fee for the affected service. They have to be requested; they are rarely paid out automatically. And they are often the only remedy the contract allows, which means a credit of a few euros may be all you get for an outage that cost you far more. For a sense of that gap, see the cost of website downtime. An SLA tells you how seriously a provider takes reliability; it is not insurance.
How to pick an uptime target
Whether you are promising uptime to customers or setting an internal goal, the same steps apply.
- Measure first. Run external checks for a month or two and see what you achieve today. Put the downtime you had into the calculator, for example 90 minutes in a month, and it tells you the highest common SLA you stayed within.
- Count your dependencies. If your site needs your server, your database and a payment provider, each with its own failures, your uptime cannot be higher than all of them combined. How to calculate uptime percentage shows how to combine them.
- Match the target to how you respond. If nobody is on call at night, a monthly budget of 22 minutes is a promise you cannot keep.
- Promise less than you aim for. Keep an internal goal one step above the external promise, for example aim for 99.95% and promise 99.9%. SLA vs SLO vs SLI explains how the two fit together.
- Write down what counts. Components covered, maintenance rules, minimum outage length and where it is measured, before anyone signs.
For most business websites and SaaS products, 99.9% a month is a realistic promise. Go higher only when your history and your setup support it.
Measuring uptime against your target
Uptimecheck’s uptime monitoring checks your sites and APIs every minute from Stockholm, Sweden. A failed check is confirmed by a second check about 10 seconds later, so a single blip never counts as an outage, and a confirmed outage is counted from the first failed check.
- Reports against an SLA target. Choose this month, last month or the last 7, 30 or 90 days, and a target from 99% to 99.99%. Monitors below the target are marked, and the report downloads as CSV. See how to run reports against an SLA target.
- Maintenance that doesn’t count. During a maintenance window the monitors aren’t checked, no alerts go out and the time doesn’t count against uptime.
- The same numbers for customers. A status page shows 90 days of uptime history for the components you choose.
Know how much of this month’s budget is left before your customers ask. Try Uptimecheck free for 14 days, no credit card required.