99.99% uptime: how much downtime is that?

Four nines allows 52 minutes of downtime a year, or 4 minutes 23 seconds a month. What that budget means for deploys, on-call and monitoring, and when 99.99% is worth promising.

· 5 min read · By the Spot Downtime team

99.99% uptime allows 52 minutes and 36 seconds of downtime a year. That's 4 minutes 23 seconds a month, about a minute a week, or under 9 seconds a day. “Four nines” is a common target for payment systems, login providers and cloud infrastructure, and a common line in sales contracts that the team behind them can't actually keep.

99.99% uptime in minutes

PeriodAllowed downtime at 99.99%
Per day8.6 seconds
Per week1 minute
Per month4 minutes 23 seconds
Per quarter13 minutes 9 seconds
Per year52 minutes 36 seconds

A month here is an average month (one twelfth of a 365.25-day year). Next to its neighbours:

UptimeDowntime per monthDowntime per year
99.9% (three nines)43m 50s8h 45m 58s
99.95%21m 55s4h 22m 59s
99.99% (four nines)4m 23s52m 36s
99.999% (five nines)26s5m 16s

Going from 99.9% to 99.99% cuts the budget by a factor of ten: from a long lunch break of downtime a month to the time it takes to make a coffee.

Uptime calculatorConvert any uptime percentage to allowed downtime, or work backwards from the downtime you can afford.

What 4 minutes a month means in practice

Humans are too slow

A typical on-call response looks like this: the alert fires, someone wakes up, finds their laptop, logs in, reads the dashboards and works out what's wrong. Fifteen minutes is a good result. At 99.99%, that one incident uses more than three months of budget. A single one-hour outage uses the whole year's.

So four nines can't rely on people fixing things. It needs systems that recover by themselves: health checks that take a bad server out of the load balancer, databases that fail over to a replica, and capacity in more than one zone or region.

Deploys must be invisible

Twenty deploys a month, each with a 15-second restart, add up to 5 minutes, more than the entire budget, before anything has actually gone wrong. 99.99% needs rolling or blue-green deploys with no gap in service, and database migrations that don't lock busy tables.

Your dependencies multiply

If your product needs three services and each one is 99.99% available, together they're about 99.97%. That's already about 13 minutes a month. Your DNS host, CDN, database, payment provider and cloud region all count. Cloud providers generally only promise four nines for deployments spread across several availability zones; a single virtual machine is usually covered by a lower SLA.

Rule of thumb

You can't be more available than the things you depend on. To promise 99.99%, every critical dependency must be better than 99.99%, or redundant.

You can't measure four nines with slow checks

If you check your site every 5 minutes, an outage can be in progress for nearly 5 minutes before the first failed check. At 99.99%, that detection delay alone is more than the whole monthly budget, and shorter outages fall between checks and never get counted.

  • Check every 30 seconds from several locations, so you see outages within a minute and can tell a real outage from one network's hiccup.
  • Check that it works, not just that it answers. A page returning 200 with an error message on it is down for your users. Use keyword checks, or API checks that exercise the database.
  • Agree what “down” means. Is one failing region an outage? Is a 10-second response? Write it down before you promise a number.

Error budgets: using the 4 minutes well

The useful way to think about 99.99% is as a budget to spend, not a score. Teams that run to four nines usually agree a simple policy:

  1. Track downtime against the month's 4 minutes 23 seconds.
  2. While there's budget left, ship features as usual.
  3. When it's spent, pause risky changes and work on reliability until the next period.

This turns “how reliable should we be?” from an argument into a number everyone can see.

Should you promise 99.99%?

Four nines is worth the engineering when downtime is very expensive, or when other businesses build on top of you:

  • Payments, checkout and authentication, where every minute down loses money or locks users out.
  • APIs and infrastructure that other companies' products depend on.
  • Anything where your contract pays meaningful service credits.

For most SaaS products, internal tools and content sites, 99.9% or 99.95% is honest and achievable. A smaller number you always meet earns more trust than four nines you miss every quarter. Measure for a few months first, and promise less than you achieve.

The short version

  • 99.99% uptime allows 52m 36s of downtime a year, 4m 23s a month, about 1 minute a week.
  • That's too short for humans to fix things: it needs automatic failover and zero-downtime deploys.
  • Dependencies multiply, so every critical one must be more available than you, or redundant.
  • Measuring it needs checks every 30 seconds or so; 5-minute checks can miss the whole budget.

For three nines and the general maths of uptime, see what 99.9% uptime really means. To find out where you actually stand, start monitoring for free. Pro checks every 30 seconds from several regions, and a public status page shows your uptime history to customers.

Get new guides by email

Practical posts on monitoring, outages, error fixes and on-call, when they're published. No spam.

Unsubscribe from any email with one click. See our privacy policy.

Keep reading