Uptime and SLA calculator

What the nines mean

Availability is quoted as a percentage of time a service is up. Each additional nine cuts allowed downtime by a factor of ten:

99% (two nines) — 3.65 days a year, 7.2 hours a month
99.9% (three nines) — 8.77 hours a year, 43.8 minutes a month
99.99% (four nines) — 52.6 minutes a year, 4.4 minutes a month
99.999% (five nines) — 5.26 minutes a year, 26 seconds a month

Written as percentages these look almost identical. Written as time they are wildly different, which is exactly why SLAs are argued over in minutes rather than percentages.

The cost curve

Each nine is roughly ten times harder than the last. Three nines is achievable with good monitoring, sensible deployment practice and a single well-run region. Four nines needs redundancy at every layer, automated failover and no single points of failure. Five nines effectively rules out manual intervention entirely — 26 seconds a month is less time than it takes a human to read an alert.

Worth noting: at five nines, the deployment process itself becomes the main risk. Most outages at that tier are self-inflicted by changes, not caused by hardware.

Read the definition, not the number

An SLA percentage is meaningless without knowing what counts as downtime. The details do the real work:

Scheduled maintenance — very often excluded entirely, so a "99.99%" service may be down for hours legitimately.
Measurement window — monthly resets are far more forgiving than annual. An outage that breaches a monthly SLA may be invisible in an annual average.
What counts as "down" — total unavailability, or degraded performance? Partial outage affecting some users? Elevated error rates?
Who measures — the provider's own monitoring usually decides, and it often checks from inside their network.
The remedy — typically a service credit worth a fraction of the monthly fee, not compensation for your losses.

Composite availability

Chained dependencies multiply. A service depending on three components each at 99.9% has a best case of 0.999³ = 99.7% — nearly three times the downtime of any single part.

Redundancy works the other way. Two independent components at 99% each, either of which can serve traffic, give 1 − (0.01 × 0.01) = 99.99%. This is the whole argument for redundant design, with the caveat that the components must fail independently. Two servers in the same rack on the same power feed are not independent.

Error budgets

A more useful way to hold the number is as a budget rather than a target. If your SLA allows 43 minutes of downtime a month, that is 43 minutes you are permitted to spend. Spend it on risky deployments and you ship faster; run out and you freeze changes until the window resets.

This reframing, standard in site reliability engineering, turns availability from a stick into a shared currency between the people who want to ship and the people who carry the pager.

Common questions

How much downtime does 99.9% allow?

About 8 hours 45 minutes per year, or roughly 43.8 minutes in a 30-day month. Three nines sounds close to four, but four nines allows only 4.4 minutes a month — ten times less.

What is the difference between 99.9% and 99.99%?

A factor of ten in allowed downtime, and usually far more than a factor of ten in cost. Three nines is achievable with good practice in one region; four needs redundancy everywhere and automated failover.

Does scheduled maintenance count against an SLA?

Usually not. Most SLAs exclude announced maintenance windows, so a service can be legitimately unavailable for hours without breaching its stated availability. Always read that clause before comparing providers.

How do I combine the availability of several services?

For dependencies in series, multiply them — three components at 99.9% give 99.7% overall. For redundant components, multiply the failure probabilities instead, so two independent 99% systems give 99.99%.