Uptime percentage calculator

The nines are worth knowing by heart because each one costs roughly ten times the last. Three nines allows 8 hours 46 minutes a year and is achievable with careful operations on a single system. Four nines allows 52 minutes and needs redundancy — that is less than one unplanned reboot. Five nines allows 5 minutes 15 seconds across a whole year, which is less than most systems take to restart, so it can only be reached by never having a single point of failure at all.

The monthly figure matters more than the annual one, because most SLAs are measured and credited monthly. Three nines monthly is 43 minutes, and one bad afternoon breaches it even though the annual figure still looks fine. An SLA that is measured monthly and quoted annually is measuring the stricter thing.

Read what the SLA excludes, because it is usually where the number comes from. Scheduled maintenance is almost always excluded, and sometimes so are provider-network problems, DDoS and anything the customer’s own configuration caused. That is why a provider can report 99.99% for a month in which you were down for two hours — the contractual figure and the figure your users experienced are genuinely different numbers, and only one of them shows up in a credit.

One compounding trap: dependencies multiply. A service at 99.9% that depends on three other services each at 99.9% has a realistic ceiling near 99.6%, because any one of them failing takes you down.