Reading uptime numbers: what 99 percent, 99.9 percent, and 99.99 percent really mean
A service that promises 99 percent uptime sounds rock solid. The math says it can be down for more than three days a year. Here is how to read uptime numbers without being fooled.
What uptime means in simple terms
Uptime is the share of time a service is working as expected. It is usually shown as a percentage measured over a fixed window such as a month, a quarter, or a year. The opposite, the time it is not working, is called downtime.
If you run a service for a thousand hours and it is down for ten of those hours, your uptime is 99 percent and your downtime is one percent.
The nines, in plain numbers
In the cloud world, people talk about uptime in terms of nines. Each extra nine cuts the allowed downtime by a factor of ten.
Here is how the most common uptime numbers translate to real time per year. These are rounded for easy reading.
- 99 percent uptime means about 3 days and 15 hours of downtime per year.
- 99.5 percent uptime means about 1 day and 19 hours per year.
- 99.9 percent uptime, or three nines, means about 8 hours and 45 minutes per year.
- 99.95 percent uptime means about 4 hours and 22 minutes per year.
- 99.99 percent uptime, or four nines, means about 52 minutes per year.
- 99.999 percent uptime, or five nines, means about 5 minutes per year.
When you see a service offer 99.9 percent uptime, that is almost nine hours per year where the service is allowed to be down. For some apps that is fine. For a bank, a phone network, or a hospital system, it is not.
Monthly versus yearly uptime
The window matters. Many cloud providers measure uptime monthly. A month has about 720 hours. So 99.9 percent monthly uptime allows about 43 minutes of downtime per month, not per year.
Always check whether a service level agreement, or SLA, is monthly or yearly. A monthly target is more generous to the provider because they can have a few bad months and still hit the yearly average.
Why nobody hits 100 percent uptime
Some downtime is unavoidable. Hardware fails. Network providers have outages. Cloud regions go offline. Software updates introduce bugs. Power gets cut. Even with perfect engineering, the supply chain that keeps a service running has many parts that can fail.
Engineers reduce the impact by running services in many places at once, having backup systems ready, and rolling out changes carefully. Those steps cost real money and they are the reason higher uptime numbers are more expensive.
What the SLA fine print usually says
Service level agreements set the rules of the game. They say what the provider promises, how they measure it, and what you get back if they miss the target. Most service credits are small, usually a partial refund on your bill for the month.
A few things to look for in any SLA:
- What counts as downtime. Some SLAs only count complete outages. Others count slow responses or partial failures.
- Scheduled maintenance. Many SLAs exclude planned maintenance windows from the downtime total.
- Refund or credit amount. Most credits are small. Read what the maximum credit can be.
- How you claim a credit. Often you have to file a claim within a fixed number of days, with logs and timestamps.
Treat the SLA number as a target the provider is willing to be measured against, not a guarantee of personal experience.
Your uptime is not the same as the provider's uptime
Even if the provider has 99.99 percent uptime, your real uptime can be worse. Your internet, your DNS, your code, your third party integrations, and your team's deploys all add their own risk. If five systems each have 99.9 percent uptime, your combined uptime is roughly 99.5 percent.
That is why teams who care about uptime invest in monitoring, retries, fallbacks, and clear status pages. Each layer adds back a little of the reliability they lost by depending on so many parts.
How to use uptime numbers when choosing tools
Do not compare uptime percentages in isolation. Ask:
- How does the provider measure uptime, and over what window?
- What is excluded from the count?
- What is the actual public outage history, not the marketing number?
- Is there an SLA, and what does it really pay out?
- Does the provider publish a status page and a postmortem after incidents?
A provider that publishes detailed postmortems and a real status page is often more reliable in practice than one that promises a higher number with no transparency.