SLA Uptime Formula: How to Calculate Uptime Percentage

The SLA uptime formula, worked examples, a downtime table for 99% to 99.999%, and the traps that make the number mislead: exclusions, composites, partial outages.

The SLA uptime formula is Uptime % = ((Total time − Downtime) ÷ Total time) × 100, with both values in the same unit and measured over the same window, usually a calendar month. OpManager’s guide gives the same standard form. The arithmetic takes ten seconds. The hard part is deciding what counts as downtime, which window you measure, and who does the measuring. This guide covers all of it, with worked examples and a lookup table.

Uptime Beacon
Measure Uptime With Fewer False Alarms
Uptime Beacon monitors endpoints, ports and hosts from five regions, opening incidents after repeated failures to help avoid network-blip alerts.
Explore uptime monitoring

TL;DR

  • Uptime % = (Total time − Downtime) ÷ Total time × 100. Use one unit, one window.
  • In a 30-day month, 99.9% allows 43.2 minutes of downtime, 99.99% allows about 4.3 minutes, and 99.999% allows about 26 seconds.
  • You can also compute availability as MTBF ÷ (MTBF + MTTR) × 100, which shows how failure frequency and recovery speed each move the number.
  • The definition of downtime matters more than the percentage. Contracts often exclude maintenance, partial degradation and customer-caused errors.
  • Multiplying SLAs of dependent services gives a ceiling, not a forecast.

Table of Contents

The formula, step by step

Write it out once and you will never need a calculator page again:

  1. Pick the window. A calendar month is the common choice for SLAs and billing credits. A 30-day month is 43,200 minutes, a 31-day month is 44,640.
  2. Convert everything to one unit. Minutes work for most teams. Seconds are better once you reach four or five nines.
  3. Add up downtime inside that window, using your agreed definition of an outage.
  4. Subtract and divide. (Total − Downtime) ÷ Total.
  5. Multiply by 100.

Worked example: one bad hour

Your API was unavailable for 60 minutes in a 30-day month.

  • Total time: 30 × 24 × 60 = 43,200 minutes
  • Uptime: (43,200 − 60) ÷ 43,200 = 0.99861
  • Result: 99.86%

One hour of downtime is enough to miss a 99.9% target, because that target leaves only 43.2 minutes a month. This surprises people who think of 99.9% as “basically always up.”

Worked example: several short incidents

Three incidents in a month: 7 minutes, 12 minutes and 4 minutes. Total downtime is 23 minutes.

(43,200 − 23) ÷ 43,200 × 100 = 99.947%

This passes 99.9% with about 20 minutes to spare and misses 99.95%, which allows 21.6 minutes. Round carefully. An SLA clause should say how many decimals count, since 99.949% and 99.95% can sit on opposite sides of a credit tier.

Monthly downtime compared with two SLA targets

Worked example: request-based availability

Some providers count failed requests rather than minutes. If your service handled 1,000,000 requests and 500 failed, availability is (1,000,000 − 500) ÷ 1,000,000 × 100 = 99.95%. Request-based measures reflect user experience better for busy services, because a one-minute outage at 3 a.m. and one at noon are not equal. They are also harder to audit from the outside, since you need the provider’s request logs.

How much downtime each “nine” allows

The figures below use a 30-day month (43,200 minutes) and a 365-day year (525,600 minutes). The 99.9%, 99.99% and 99.999% rows match the allowances OpManager lists; the other rows follow from the same formula.

Uptime target Downtime per 30-day month Downtime per year
99% 7.2 hours 3.65 days
99.5% 3.6 hours 1.83 days
99.9% 43.2 minutes 8.76 hours
99.95% 21.6 minutes 4.38 hours
99.99% 4.32 minutes 52.6 minutes
99.999% 26 seconds 5.26 minutes

Two things to notice. Each extra nine cuts the allowance by a factor of ten. And around 99.99% the budget is smaller than the time it takes a human to open a laptop, read an alert and type a command. At that level recovery has to be automatic, which is a very different engineering project than 99.9%.

To reverse the calculation, take the allowed downtime in minutes, divide by total minutes, and subtract from 1. Or go forward: allowed downtime = total time × (1 − target).

The second formula: MTBF and MTTR

The time-based formula tells you what happened. The reliability formula tells you what to change. Availability can be written as MTBF ÷ (MTBF + MTTR) × 100, where MTBF is the mean time between failures and MTTR is the mean time to repair.

Google Cloud uses this in its Memorystore four-nines write-up: with an MTBF of 30 days and an MTTR of 20 minutes, availability comes to 99.95%. Check it: 30 days is 43,200 minutes, and 43,200 ÷ 43,220 = 99.954%.

The formula shows two levers. Failing less often raises MTBF. Recovering faster lowers MTTR. For a small team, MTTR is usually the cheaper lever: faster detection, a cleaner runbook, a rollback that takes one command. Be aware that “MTTR” stands for several different things depending on who says it, and the number you plug in must match your downtime definition. We break that down in One Outage, Three Different MTTRs.

What counts as downtime is the real question

Two companies can both promise 99.99% and mean very different things. Microsoft’s guidance on reading SLAs makes the point that a figure like that means little if downtime only covers total collapse, because slow responses and partial failures can still stop users from finishing a task. It suggests definitions tied to user experience, such as failed request rate.

Before you sign or publish an SLA, pin down these items:

  • Measurement interval. AWS, for example, defines availability per five-minute interval in its ECR SLA, and monthly uptime is the average of those intervals. An interval with no requests is treated as fully available. A short outage inside an interval may therefore count differently from the same outage measured by the second.
  • What failure means. HTTP 5xx? Timeouts above two seconds? Login failures? Failed checkout? Choose the one that matches what a customer feels.
  • Exclusions. Scheduled maintenance, customer-caused errors and force majeure are commonly carved out. Read them. A generous exclusion list can make a 99.99% promise nearly unfalsifiable.
  • Who measures. Provider-side logs, or an independent monitor? If you are the one promising uptime, an external measurement protects you in a dispute.
  • The remedy. Most SLAs pay service credits, not cash. Amazon’s WorkMail SLA lists tiers: a 10% credit when monthly uptime falls below 99.9% but stays at or above 99.0%, and 25% between 99.0% and 95.0%. The credit is a share of your bill, not compensation for your lost revenue.

Composite SLAs: why multiplying is only a ceiling

If your product depends on a database at 99.9% and a cloud region at 99.9%, the textbook move is to multiply: 0.999 × 0.999 = 99.8%. With three such dependencies you get 99.7%, which is about 2.2 hours a month.

Treat that as an upper bound, not a prediction. Microsoft notes that multiplication assumes failures are independent, which distributed systems rarely honor, and recommends failure mode analysis instead. A shared network, a shared deploy pipeline or a shared DNS provider can take down several “independent” services at once. In the other direction, retries, caches and failover can raise your real availability above the product of your parts.

The practical rule: never promise customers a higher SLA than the multiplied figure of your hard dependencies unless you can show where redundancy covers the gap.

From SLA to SLO and error budget

An SLA is the promise with consequences. Inside the team, you usually run a stricter objective, an SLO, so you have room to react before the contract is at risk. Google’s SRE guidance frames the remainder as an error budget: for a 99.9% target, 0.1% is your tolerance for unreliability. Spend it on releases and experiments, and once it is gone, reliability work takes priority over new features.

A simple way to apply it:

  1. Set the public SLA (say 99.9%).
  2. Set an internal SLO slightly above it (say 99.95%).
  3. Track budget burned each week: downtime minutes so far ÷ allowed minutes.
  4. Agree in advance what happens at 50%, 75% and 100% burned.

For how to choose targets that hold up, see Service Level Objectives: How to Set Ones That Work.

Our perspective: bad data breaks a good formula

The formula is only as honest as its downtime input. We see two ways it goes wrong in practice, and both come from the monitoring layer, not the math.

First, false downtime. A single failed check from one location can be a network blip between the probe and your server, not an outage. If a monitor counts that as downtime, your computed uptime drops for no real reason, and an on-call engineer gets woken up for it. This is why Uptime Beacon opens an incident only after several failures in a row, and on paid plans confirms from a second location before it does. The incident record then reflects a failure your users could plausibly have felt.

Second, resolution. If you check every five minutes, as the Free plan does, you cannot measure downtime more finely than five minutes. One failed check can stand for up to five minutes of unknown. That is fine for a 99.9% internal target, where the monthly budget is 43.2 minutes. It is thin for 99.99%, where the whole month’s budget is 4.32 minutes. Match the check interval to the target before you quote the percentage.

We would also add a point about the people. Our founders have more than 30 years combined running production in banking, fintech, telecom and SaaS, and the lesson that repeats is that teams argue about the nines and neglect the definition. Write down what “down” means, measure it from outside your own infrastructure, and keep the evidence per region so a disputed incident can be settled in minutes.

— Aaditya Parashar

Measuring your own uptime in practice

You do not need a spreadsheet of stopwatch times. A workable setup:

  • Monitor what users hit. Check the real endpoint, not just the home page. For APIs, a health route that exercises the database tells you more than a static 200. Our guides on API uptime monitoring cover what to assert and when to alert.
  • Check from more than one place. A failure seen from a single region may be local routing.
  • Separate planned from unplanned. If your SLA excludes scheduled maintenance, record maintenance windows so the calculation can exclude them with a paper trail.
  • Compute monthly, review weekly. Monthly for the contract, weekly for the budget burn.
  • Keep incident records. Start time, end time, cause and affected regions are what you need to defend or correct a figure.

If you want this off your plate, Uptime Beacon’s Free plan needs no card and gives you 5 monitors checked every 5 minutes from EU Central, one alert channel (email or Slack) and one status page. Paid plans start at €14.99 per month for Starter, with Pro at €39.99 and Business at €89.99 per month, and add second-location confirmation of failures. If you also run scheduled jobs or an API product, our SaaS monitoring page shows how heartbeat and endpoint checks fit together. Results are stored on a control plane and database in the EU.

On every plan, including Free, you can also connect an AI assistant such as Claude, Cursor or ChatGPT over MCP and ask it to read check history and diagnose a failing monitor region by region, which makes month-end uptime questions quicker to answer.

Measuring your own uptime in practice — overview diagram

Common mistakes when calculating uptime

Mixing units. Dividing 43 minutes by a month expressed in hours gives nonsense. Convert first.

Using the wrong month. February has 40,320 minutes, a 31-day month 44,640. Calculators usually assume 30 days, so the allowance for a given nine differs slightly. State your assumption in the SLA.

Averaging percentages. If January was 99.95% and February 99.80%, the two-month figure is not the plain average unless the months have equal minutes. Add the downtime and total time, then divide.

Ignoring partial outages. A checkout that fails for 30% of users while the home page works is still an outage to those users. Decide whether your definition counts a degraded state, and at what threshold.

Quoting a number without a window. “99.9%” alone says nothing. 99.9% per month and 99.9% per year allow very different worst cases, since the yearly figure lets you burn 8.76 hours in a single day.

Chasing nines you cannot staff. The jump from 99.9% to 99.99% is not a configuration change. It usually means redundancy, automated failover and a pager that is answered in seconds.

FAQ

What does a 99.9% uptime SLA mean?

It means the service may be unavailable for at most about 43.2 minutes in a 30-day month, or 8.76 hours in a year, before the provider is in breach. What happens after a breach depends on the contract, and it is typically a service credit.

How do I calculate uptime percentage from downtime minutes?

Subtract the downtime minutes from the total minutes in the period, divide by the total, and multiply by 100. For 90 minutes of downtime in a 30-day month: (43,200 − 90) ÷ 43,200 × 100 = 99.79%.

How much downtime is 99.999% uptime?

About 26 seconds per 30-day month and roughly 5.26 minutes per year. At that level, human response is too slow, so recovery has to be automated, and the measurement itself needs to be finer than the check interval most monitors use.

Is availability the same as uptime?

In most SLAs the two terms are used interchangeably, both as the share of time a service works. Strictly, availability is whether a system can fulfill its function at a given point in time, so how you define “function” decides the figure. Check the contract’s definition rather than relying on the label.

Should I include scheduled maintenance in downtime?

That is a contract choice. Many SLAs exclude announced maintenance, while stricter customers push for it to count. Whichever you choose, write it down and log maintenance windows so the exclusion can be shown, not just asserted.

Sources

Put your site on it. Sleep through the blips.

Add your first monitor now. If your site is fine, you will not hear from us.

Monitor it free
Sign up with GoogleOpening Google…
Free plan, no cardSign up with Google or email