99.9% uptime: how many minutes of downtime are actually allowed?

A report says "99.9% uptime this month" and it looks like a good score. But 99.9% of what, calculated how, and how many minutes of downtime does it actually cover? The answer varies considerably depending on the threshold and the reference period—and that's the question you need answered before signing a hosting contract or accepting a report.

The formula that (almost) everyone uses

The foundation is always the same: Uptime % = ((total time − downtime) / total time) × 100, as SLA calculators in the industry report. Working backwards, from measured downtime to uptime percentage: if the service was down for 1 hour in a 30-day month (720 hours), your uptime is ((720 − 1) / 720) × 100 = 99.86%.

To convert a percentage threshold into permitted minutes, the formula is reversed: permitted downtime = period duration × (1 − SLA/100).

The threshold table

Applying this formula to a 30-day month and a 365-day year produces the table that most industry calculators report consistently:

SLA Threshold	Allowed downtime per month	Allowed downtime per year
99.0%	7 hours 18 min	3 days 15 hours
99.5%	3 hours 39 min	1 day 19 hours
99.9%	43 minutes	8 hours 45 min
99.95%	21 minutes	4 hours 22 min
99.99%	4 minutes	52 minutes
99.999%	26 seconds	5 minutes

These figures align with those published across the industry: from 99.0% with 3 days 15 hours of allowed downtime per year (7 hours 18 minutes per month) up to 99.999% with 5 minutes per year and 26 seconds per month. Each additional nine reduces permitted downtime by a factor of ten: the difference between 99.9% and 99.99% isn't "a decimal point", it's the difference between 43 minutes and 4 minutes per month.

Monthly or annual: the period changes everything

The same 99.9% measured on a monthly or annual basis isn't equivalent in practice. Measuring 99.9% on an annual basis allows a single outage of nearly 9 hours to pass, provided the rest of the year stays clean. Monthly windows carry more weight, and that's what most providers measure. Before comparing two providers, check which time window they use to calculate the percentage: a 99.9% "annual" figure allows far more latitude than a 99.9% "monthly" one.

How the number in the report is actually calculated

The procedure that follows (and which you can repeat yourself) is: export the month's monitoring logs, sum all incidents in minutes, divide by the month's total minutes and subtract from 100. A concrete example: in a month of 40,320 minutes, with 3 incidents totalling 52 minutes of downtime, the calculation is ((40,320 - 52) / 40,320) × 100 = 99.871%—below 99.9%, so below SLA threshold.

What to ask if the numbers don't match

If the provider's report doesn't match what you measured with an external monitor, three things to ask immediately:

• Raw logs, not just the final number: logs from your external monitoring tool showing the specific periods and durations of downtime.
• Contractual exclusions: scheduled maintenance, acts of God, network issues outside the provider's control often don't count as downtime.
• What counts as "down": the subtlety lies in what gets considered downtime—a partial degradation, a single region failing, or a slow but responsive endpoint may or may not be counted depending on how the agreement defines availability.

On credits for SLA breach, the common practice is a percentage scale of the monthly fee, not compensation for actual damages: a service credit is a common way providers reimburse customers for an uptime percentage violation in an SLA, generally on a tiered scale—for example "5% of monthly cost for every hour out of SLA". On a plan costing just a few pounds per month, the credit covers the disruption only on paper: read it as a process guarantee, not as insurance against economic loss.