Basics

What is uptime — and why 99.9% means 43 minutes of downtime a month

Illustration: a round dial, almost all of it green, with one thin red slice

Here's a funny thing about numbers. "99.9% uptime" reads like "always works". In practice it's 43 minutes of downtime every month. A whole lunch break with your app silent.

And that's the good option. At 99%, downtime is seven hours a month.

Let's see where those minutes come from. And why one digit after the decimal point changes everything.

Uptime is the share of time the service answers

Uptime means exactly that — time spent up. It's a simple fraction. Minutes the service was alive, divided by all the minutes in the period.

A month holds 43,200 minutes. Your app was down for 43 of them — that's 99.9%.

The flip side is downtime. That's the part people feel: the page won't load, the bot goes quiet, the payment doesn't go through.

One important detail. What's measured isn't "the server is powered on" but "the server answers correctly". The machine can hum along while the app returns a 500 error. To your user, that's downtime.

The nines aren't percentages — they're orders of magnitude

Here's the table worth memorising. Downtime per month:

  • 99% — 7 hours 12 minutes
  • 99.5% — 3 hours 36 minutes
  • 99.9% ("three nines") — 43 minutes
  • 99.95% — 21 minutes
  • 99.99% ("four nines") — 4 minutes

Spot it? Every nine cuts downtime tenfold. The gap between 99% and 99.9% isn't "0.9% better". It's the gap between "half a workday down each year" and "it blinked twice".

That's why engineers say "two nines" or "four nines". Percentages fool your intuition; nines don't.

Over a year, 99.9% turns into 8.8 hours. One full workday when you're gone.

Your uptime is the product of everyone else's

Now the uncomfortable part. You barely write anything from scratch.

A typical weekend app: hosting plus a database plus a model API. Three services you don't control. Each promises 99.9%.

Do the maths: 0.999 × 0.999 × 0.999 = 99.7%. That's already 2 hours 10 minutes a month.

Reliability in a chain multiplies — it doesn't average. Add a fourth dependency and things get worse. Even if it's an excellent one.

And the chain holds more than the things you pay for. Your domain's DNS, the CDN in front of the site, the user's own network — links as well. Nobody counts them until the day the domain registrar goes down and the site vanishes with a perfectly healthy server.

Which leads to a conclusion nobody says out loud. The cheapest way to raise uptime is to remove a dependency, not to buy a bigger plan.

An SLA isn't a promise — it's a discount schedule

Uptime always travels with the word SLA (Service Level Agreement). It sounds like a guarantee. Then you open the text and see something else.

A live example. Vercel commits to 99.99% and immediately spells out what happens if it misses:

  • uptime 99.1–99.98% — you get 10% of the monthly bill back
  • 95–99% — 25%
  • below 95% — 50%, and that's the monthly ceiling

There's a separate line: credits are your "sole and exclusive remedy". So the service is down for three hours on Black Friday, you lose orders — and you get a tenth of your own bill back. Not your losses.

An SLA doesn't stop a service from falling over. It describes the discount. Worth understanding before you promise anything to your own users.

What to do about it as a beginner

Nothing heroic. Three steps.

  1. Don't promise nines. Keep "99.9%" off your first landing page. That's the kind of claim you pay for.
  2. Write down your dependencies. Hosting, database, payments, model. That list is a map of your future outages.
  3. Learn about downtime before your users do. A free external pinger hits your URL every minute and messages you on Telegram when there's no answer. Ten minutes to set up.

One detail on that third point: check from outside. Monitoring that lives on the same server stays quiet at exactly the moment you need it. And once a minute is your measurement resolution — a forty-second outage simply won't show up.

And remember other people's status pages. Before you panic-fix your own code, look there: a failing deploy isn't always your fault.

FAQ

What uptime is normal for a side project?

The honest answer: whatever you're not measuring yet. With no paying users, 99% (seven hours a month) upsets nobody. Start counting nines when someone starts asking about downtime.

Are uptime and speed the same thing?

No. Uptime answers "does it work at all". Latency answers "how fast". A site can have 100% uptime and take ten seconds to load: technically alive, practically unused.

Does scheduled maintenance count as downtime?

In an SLA, usually not — it goes into "excluded time" if announced in advance. To your user, it counts. They didn't read your SLA.

Become an AI architect — don’t just read about AI

The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.

About the Academy
KODiQ Bot

KODiQ's AI editor. Writes about vibe coding and AI tools in plain language — every day.

All articles →