SLA Comparison Table
What each availability target actually allows, from 99% to five nines.
What you provide
Result
99.9% allows 43.2 min of downtime a month
Each extra nine costs roughly ten times what the last one did, and buys back an order of magnitude less time. The step from 99.9% to 99.99% is the one where architecture changes: it leaves about four minutes a month, which is less than a person can notice a problem, log in and fix it, so recovery has to be automatic. Read the measurement window too. A monthly target hands back a fresh budget twelve times a year, and most agreements exclude scheduled maintenance, which is where a stated availability and an experienced one part company.
| Check | Per day | Per month | Per year |
|---|---|---|---|
| 99% | 14.4 mininfo | 7.20 hinfo | 3.65 daysinfo |
| 99.5% | 7.2 mininfo | 3.60 hinfo | 1.82 daysinfo |
| 99.9% | 1.4 mininfo | 43.2 mininfo | 8.76 hinfo |
| 99.95% | 43.2 sinfo | 21.6 mininfo | 4.38 hinfo |
| 99.99% | 8.6 sinfo | 4.3 mininfo | 52.6 mininfo |
| 99.999% | 864 msinfo | 25.9 sinfo | 5.3 mininfo |
- 99%3.65 days a year
- 7.20 h / month
- 99.5%1.82 days a year
- 3.60 h / month
- 99.9%8.76 h a year
- 43.2 min / month
- 99.95%4.38 h a year
- 21.6 min / month
- 99.99%52.6 min a year
- 4.3 min / month
- 99.999%5.3 min a year
- 25.9 s / month
What this cannot tell you
- A reference table computed from percentages. It does not measure any service and knows nothing about your actual availability.
- Every agreement defines downtime differently. Scheduled maintenance, partial degradation and the measurement window all change what a percentage means in practice, and none of that is in the arithmetic.
Take this with you
How this calculation works
Each availability percentage is converted into the downtime it permits over a day, a month and a year. Laid out together, the table makes the shape of the problem visible in a way a single figure does not: each additional nine cuts the allowance by a factor of ten, and the cost of achieving it goes up by roughly the same factor. The month is taken as thirty days, which is the convention almost every agreement uses.
What the results mean
- Per month
- The usual contractual window, and the one that matters most. A monthly target hands back a fresh budget twelve times a year, so a bad month does not have to be paid for later.
- Per year
- Useful for comparing against your own incident history. Add up last year real outages and see which row you were actually on.
- The step between rows
- Where the architecture changes. Getting from 99.9% to 99.99% leaves about four minutes a month, which is less time than a person needs to notice and log in.
Common problems and fixes
- We are told we have 99.99% but outages feel more frequent
- Check what the agreement counts. Most exclude scheduled maintenance entirely, many measure availability at a component rather than at the service a user experiences, and partial degradation often counts as up. A service that responds with errors for a third of requests may still be inside its target while feeling broken to everyone using it.
- A vendor quotes uptime across multiple dependent services
- Multiply, do not average. If your service needs three components that each promise 99.9%, the combined availability is 0.999 cubed, which is 99.7% and about double the downtime. Serial dependencies compound quickly, and this is the arithmetic behind the case for removing them rather than adding more nines to each one.
Frequently asked questions
How much downtime is 99.9%?
About 43 minutes and 12 seconds a month, or 8 hours and 45 minutes a year. It is enough to absorb one bad incident a month with a person responding manually, which is why it is the most commonly offered target.
Why is five nines so expensive?
It leaves roughly 26 seconds of downtime a month, which is less than a deployment, a failover or a person reacting to a page. Everything has to be redundant and every recovery has to be automatic, including the parts that fail rarely enough that nobody has tested them. The cost is not the hardware, it is the engineering discipline needed to never rely on a human being awake.
What does an SLA actually pay out when it is missed?
Usually a service credit worth a percentage of that month fee, and you generally have to claim it. It is almost never related to what the outage cost you. The SLA is best read as a statement of intent and a threshold for escalation, not as insurance.
Put this on your own site
Free to embed, no attribution required beyond the source link the frame carries itself. It runs entirely in your visitor's browser, sets no cookies and loads no third-party script.
<iframe src="https://runthetests.com/embed/sla-comparison/" width="100%" height="560" style="border:1px solid #e5e5e5;border-radius:8px" title="SLA Comparison Table" loading="lazy"></iframe>
Preview it at https://runthetests.com/embed/sla-comparison/. Embedded pages are marked noindex, so yours stays the canonical copy — not this one.