How much downtime is 99.9% uptime?

By Dustin Dasal

Published October 10, 2026. Updated October 10, 2026.

This guide turns an uptime percent into the hours and minutes you can put in a client agreement. If you need a percent that isn't in the table, you can add it to the page address, as explained below.

99.9% uptime allows 8 hours 45 minutes 36 seconds of downtime in a 365-day year, 43 minutes 12 seconds in a 30-day month, and 1 minute 26 seconds in a day. Each extra nine cuts that allowance by ten. The formula is period length times (1 minus the uptime percent).

Uptime here means the share of a period when the service counts as up. Availability is that share written as a percent. An SLA is the promise in a contract. An SLO is the internal target you run the service toward. The number you can defend is the one your checker actually measured.

What is the downtime formula?

Downtime is the period length multiplied by the failure budget. A 99.9% target leaves a budget of 0.1%, which is 0.001. Multiply the seconds in the period by 0.001, then round to the nearest second.

A day is 86,400 seconds. A week is 7 of those. The year in this table is 365 days, not 365.25, because 365 times 86,400 times 0.001 is exactly 31,536 seconds, which is 8 hours 45 minutes 36 seconds. A quarter is 365 divided by 4, so 91.25 days. Two month columns are shown on purpose. One is 30 days. The other is 30.44 days, which is 365.25 divided by 12, rounded to two decimals. A calendar month is neither of those. February and March do not have the same allowance.

If you need a percent that is not in the table, add it to the address, for example ?pct=99.95. The value has to be from 0 to 100 with at most four decimal places. Anything else is rejected and the standard table stays up.

Why does the month length change the answer?

People argue about 99.9% because they are using different months. A 30-day month at 99.9% is 43 minutes 12 seconds. A 30.44-day month is a little longer, about 44 minutes. A 31-day month is longer again. If the contract says "per month" and nobody says which month, you will lose the argument on a short month and win it on a long one, or the other way around.

Write the period into the sentence. "99.9% of minutes in a calendar month" is a different promise from "99.9% of a 30-day window" and from "99.9% of a rolling 365 days." Pick one and use it in the report you send the client. Do not switch definitions between the proposal and the quarterly review.

Can a one-minute check see a 20-second outage?

No. A check that runs once a minute cannot see an outage that starts and ends between two checks. A 20-second blip is real to the person who hit refresh, and it is invisible to a one-minute probe. The percent you publish is the percent of checks, or of minutes you recorded, not the percent of every instant.

That gap matters when a client compares your report with their own analytics. Their shop system may have logged errors your probe never caught. Say so before they ask. A faster check shrinks the gap and raises the bill. It does not close the gap. Nothing that samples on a timer sees every blip.

What is the difference between an SLA, an SLO, and a measurement?

An SLO is the target your team uses to decide whether the service is healthy enough. An SLA is the version of that target that has a remedy attached, usually a service credit. The measurement is the series of checks you can show. Those three numbers are not required to match, and they usually should not. The SLO can be tighter than the SLA so you notice trouble before the contract does.

Do not put a percent in a contract that your report cannot produce. If you measure one URL once a minute from outside the client's network, the SLA should say that. It should not say "the website," and it should not say "from the user's browser," unless you actually measure those things.

A credit clause needs a floor. Many MSP agreements exclude a credit below a small amount, require the client to ask within a set number of days, and cap the credit at that month's fee. The percent is the easy part. The notice rule and the cap are what you will use.

How do you write the SLA line?

Use one sentence for the target and one for what was measured. Here is a shape that stays honest:

The homepage URL listed in the schedule is targeted at 99.9% of recorded minutes in each calendar month, measured by an outside HTTPS check about once a minute. Minutes with no recorded check are left out of the percent. Announced maintenance is listed on the status page and is outside this target only when the notice in this agreement was given.

Then attach the credit, the cap, and who has to ask. If announced maintenance counts against the target, say that too. The page on maintenance windows is about how to write that exclusion so it is not a surprise at 2 a.m.

What do the common percents look like in practice?

99% sounds loose until you put it on a year: about 3 days 15 hours. A dental office that is "down half a Friday" has already used a large part of a 99% year, and almost none of a 99.9% year. 99.5% is about 1 day 19 hours a year. 99.95% is about 4 hours 23 minutes a year. 99.99% is 52 minutes 34 seconds. 99.999% is 5 minutes 15 seconds. Those last two are hosting-provider numbers. An MSP watching a brochure site on shared hosting should not sign them.

Worked example. A client has 40 sites and wants 99.9% on each homepage, calendar month, outside announced maintenance. You check once a minute. In a 30-day month the allowance is 43 minutes. Two confirmed outages of 15 minutes each, plus the two minutes you waited for confirmation on each, add up to 34 minutes, which uses most of that budget. If your agreement starts the clock at the first failed check instead of at the confirmed outage, you have even less room. Write down which clock you use.

Another example. You quote 99.9% and the client's busy month is December, which has 31 days. Your internal sheet used 30. The client is owed about a minute and a half more downtime than your sheet assumed, which is nothing, or they think you shorted them, which is a meeting. Define the month.

What should you do with the number?

Put the percent, the period, and the measurement method in the same sentence, then keep the report in those units. If you also watch the domain name, remember that a registrar problem takes the whole site and the email down together, which spends the budget in one incident. The glossary defines SLA, SLO, and the related terms if you need the short versions for a proposal.

How The Watchbill helps

The Watchbill measures the way this guide recommends you write the SLA. It checks each site about every 60 seconds from outside the client's network, and it marks a site down or up only after two consecutive minutes agree, so a single bad minute doesn't open an outage. The history behind those checks is kept for 90 days, which covers a full quarter, and it's shown in your account and on any status page you publish.

That gives you a measurement you can name in the contract: one URL, checked about once a minute, from outside. Like any check on a timer, it can't see an outage shorter than the gap between checks, so say that in the agreement too. If an outage you're explaining turns out to be a lapsed domain rather than a server, read what happens when a domain expires. You can start measuring a client's site with a free account, and the plans are on the pricing page.

Sources

1. NIST SP 800-160 Vol. 2 Rev. 1, Developing Cyber-Resilient Systems (availability as a concern, not a downtime table). Accessed October 10, 2026.

Related guides

• What happens when a domain expires
• Should maintenance count on an SLA?
• Uptime monitoring glossary for MSPs