SLA Calculator
Turn an uptime percentage into the downtime it allows per day, week, month, quarter and year. Or work back from real downtime to the uptime you hit, and combine the SLAs of the services you depend on.
Enter any value, e.g. 99.95.
Allowed downtime
Three nines- / day
- 1m 26.4s
- / week
- 10m 4.8s
- / month basis
- 43m 49.7s
- / quarter
- 2h 11m 29.2s
- / year
- 8h 45m 57s
That's your error budget: the downtime your SLO permits before you breach. How to defend it →
Reference
SLA uptime table: allowed downtime by level
The common levels from 90% to 99.999%. Click a row to load it into the calculator, or open a level for the details.
| Uptime % | Common name | / day | / week | / month | / quarter | / year |
|---|---|---|---|---|---|---|
| 90%# | One nine | 2h 24m | 16h 48m | 3d 1h 2m 54.6s | 9d 3h 8m 43.8s | 36d 12h 34m 55.2s |
| 95%# | — | 1h 12m | 8h 24m | 1d 12h 31m 27.3s | 4d 13h 34m 21.9s | 18d 6h 17m 27.6s |
| 98%# | — | 28m 48s | 3h 21m 36s | 14h 36m 34.9s | 1d 19h 49m 44.8s | 7d 7h 18m 59s |
| 99%# | Two nines | 14m 24s | 1h 40m 48s | 7h 18m 17.5s | 21h 54m 52.4s | 3d 15h 39m 29.5s |
| 99.5%# | — | 7m 12s | 50m 24s | 3h 39m 8.7s | 10h 57m 26.2s | 1d 19h 49m 44.8s |
| 99.9%# | Three nines | 1m 26.4s | 10m 4.8s | 43m 49.7s | 2h 11m 29.2s | 8h 45m 57s |
| 99.95%# | — | 43.2s | 5m 2.4s | 21m 54.9s | 1h 5m 44.6s | 4h 22m 58.5s |
| 99.99%# | Four nines | 8.6s | 1m 0.5s | 4m 23s | 13m 8.9s | 52m 35.7s |
| 99.999%# | Five nines | 0.9s | 6s | 26.3s | 1m 18.9s | 5m 15.6s |
Months and quarters are averages: 1/12 and 1/4 of a 365.2425-day year (30.44 and 91.31 days). In a real 31-day month 99.9% allows 44m 38.4s, in February 40m 19.2s. Durations in days, hours, minutes and seconds are rounded to the nearest 0.1 second. How we calculate this
The most common SLA levels, explained
Composite SLA
Combined SLA calculator
A service is only as available as everything it needs. If a request needs your load balancer, your database and a payment API, the chance that all three are up is the product of their uptimes: SLA = A × B × C. Three parts at 99.9% give 99.7002%, already worse than any one of them. Redundant copies work the other way: the service is down only when every copy is, so SLA = 1 − (1 − A) × (1 − B). That only holds if the copies fail independently; a shared database, region or deploy breaks it.
All of them needed (in series)
99.84%
1h 10m 5.9s of downtime a month, 14h 1m 10.6s a year.
Any one is enough (redundant, in parallel)
99.9999%
Rounded down to 4 decimals, so it never overstates uptime. Parallel assumes the copies fail independently.
What the nines actually mean
Each extra nine cuts the allowed downtime by a factor of 10, and usually means a large step in engineering cost. The jump from 99.9% (43m 49.7s a month) to 99.99% (4m 23s a month) typically buys you redundancy, automated failover and detection within a minute. The next step, 99.999% ("five nines"), leaves 26.3s a month: too little for a person to react, so everything has to be automatic. Whether that last fraction of a percent is worth it depends on what an outage actually costs you.
SLA vs SLO vs SLI, and your error budget
An SLI is the measured signal: for example, the percentage of successful requests. An SLO is your internal target for that signal. An SLA is the contractual promise you make to customers, usually looser than the SLO so you have headroom before a breach has financial consequences. Your error budget is what's left over: 100% minus your target. You monitor the SLI, manage to the SLO, and report against the SLA. See how uptime monitoring fits in →
How we calculate this
The core formula is downtime = (1 − uptime ÷ 100) × period. We use the average Gregorian year of 365.2425 days (365 days plus 97 leap days every 400 years), a month of 1/12 of that (30.436875 days), a quarter of 1/4 (91.310625 days), a 7-day week and a 24-hour day. Months and quarters are exact fractions of the year, so 12 months of downtime always add up to a year of it. Seconds are shown to one decimal place, and units with a zero value are dropped (so 90% over a week reads "16h 48m", not "0d 16h 48m 0s"). SLAs measured per calendar month allow a little less in February and a little more in 31-day months; each level page lists those figures too.
FAQ
Frequently asked questions
99.9% ("three nines") allows 43m 49.7s of downtime in an average month, 8h 45m 57s in a year and 1m 26.4s a day. Every extra nine cuts the allowed downtime by a factor of 10.
99.9% allows 43m 49.7s a month; 99.99% ("four nines") allows only 4m 23s. That 10 times tighter budget usually means redundancy, automated failover and detection within a minute: a large jump in engineering cost for the last 0.09%.
Your error budget is the downtime your SLO permits, 100% minus your target. At a 99.9% SLO you have a 0.1% error budget (43m 49.7s a month) to spend on incidents, risky deploys and maintenance before you breach.
An SLI is the measured signal (e.g. % of successful requests). An SLO is your internal target for it. An SLA is the contractual promise to customers, usually looser than the SLO, with penalties if you miss it. You monitor the SLI, manage to the SLO, and report against the SLA.
Uptime % = (total time − downtime) ÷ total time × 100, over a defined window. The window matters: 5 minutes of downtime is 99.9885% over an average month but only 99.6527% over a single day. Always state the measurement period alongside the number.
Multiply the uptimes of the parts that all have to work: two services at 99.9% each give 99.9% × 99.9% = 99.8001%. For redundant copies where one working copy is enough, multiply their downtime instead: two copies at 99.9% give 1 − 0.001 × 0.001 = 99.9999%, as long as they fail independently.
It depends on the contract. Many SLAs exclude maintenance that is announced in advance, and some cap how long it can be. If yours does not exclude it, every maintenance window spends error budget like an outage.
An average month: 1/12 of the average Gregorian year of 365.2425 days, so 30.436875 days. That keeps 12 months equal to one year. Real calendar months run 28 to 31 days, so 99.9% allows 40m 19.2s in February and 44m 38.4s in a 31-day month. A flat 30-day month gives 43m 12s.
From SLA target to SLA defended
A downtime budget is only as good as how fast you spot a breach.
Drumbeats checks your sites and APIs as often as every minute on the Free plan and every 30 seconds on paid plans, watches your cron jobs and workers, and alerts you the moment you start spending your error budget: before it turns into a missed SLA.
Free for up to 50 monitors · Setup in 60 seconds · No credit card required