Uptime & SLA Calculator
Allowed downtime for any number of nines, and the availability a given outage leaves you.
How it works
Availability is uptime divided by total time, and the nines are just a convenient way to say it. Each additional nine divides the allowed downtime by ten, and the jump in engineering cost between them is far steeper than the arithmetic suggests.
Three nines is 43 minutes a month — one bad deploy, or a single reboot that took longer than expected. Four nines is 4.3 minutes a month, which no human can respond to; by the time someone has been paged, opened a laptop and understood the problem, the budget is gone. Anything past three nines has to be handled by automation, and the difference between three and four is not extra effort but a different architecture.
The compounding of series components is the part that ruins plans. If a request must pass through a load balancer, an application server, a database and a cache, and each is independently 99.9% available, the chain is 0.999⁴ — 99.6%, which is 35 hours a year rather than 8.7. Availability multiplies down the chain, and every dependency you add makes the number worse. This is why redundancy exists: parallel components multiply their *unavailability* instead, which drives it toward zero.
Read the SLA's own definitions before trusting any of this. Most providers measure availability per region or per service rather than end-to-end, exclude scheduled maintenance, and require you to notice and file a claim to get the credit. And the credit is normally a percentage of that month's bill — a rounding error against what an outage actually costs you.
Common questions
How much downtime is three nines?
99.9% allows 8 hours 45 minutes a year, 43 minutes 12 seconds per 30-day month, or 1 minute 26 seconds a day. Four nines cuts each of those by a factor of ten.
Why is my system less available than its parts?
Because serial dependencies multiply. Four components at 99.9% each give 99.6% overall — four times the downtime of any single one. Every dependency in the request path lowers the ceiling.
Does redundancy fix it?
Yes, and dramatically, because parallel components multiply their unavailability instead. Two independent 99% components in parallel give 99.99% — provided they genuinely are independent, which shared power, shared network and shared configuration all quietly break.
What is an error budget?
The downtime the SLA permits, treated as a resource to be spent. If you are well inside budget you can afford to ship faster and take more risk; if you have burnt it, you stop shipping features and fix reliability. It turns availability from an argument into a number.