# 99.9% uptime sounds great until you do the math: it allows 8h 46m of downtime a year, and your real number is probably lower — retained record

> Retained reddit record from reddit:sysadmin.

Canonical: https://highsignal.app/data/records/c28ebdc0731df612

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:sysadmin)  published 2026-10-14 00:20 ingested 2026-10-15 02:41

[original source ↗](https://www.reddit.com/r/sysadmin/comments/1x5d60q/999_uptime_sounds_great_until_you_do_the_math_it/)

## retained text



```
I was reviewing an SLA last week and realized I’d never actually worked out what the nines mean in time. Here’s the table:

|Uptime|Per year|Per month|Per week|
|:-|:-|:-|:-|
|99%|3d 15h 36m|7h 18m|1h 41m|
|99.9%|8h 46m|43m 50s|10m 5s|
|99.95%|4h 23m|21m 55s|5m 2s|
|99.99%|52m 36s|4m 23s|1m 1s|
|99.999%|5m 16s|26s|6s|

8h 46m a year sounds generous. Then I thought about how we actually *measure* it, and I’m pretty sure our real uptime is lower than the dashboard says. Here’s why:

**1. Your check interval hides outages.** If you check every 5 minutes, a 4-minute outage can fall between two checks and never get recorded. Then add 2–3 failed checks before alerting to avoid false positives, and a “5-minute” monitor can take 10–15 minutes to notice anything.

**2. You’re checking from one place.** If the monitor runs in the same region or datacenter as the app, it can’t see a DNS, CDN, or routing problem that’s taking you down for half your users. It can also go down in the same incident as the app.

**3. “Up” means HTTP 200 on the homepage.** That cached homepage loads fine while login is broken, the API returns 500s, or checkout hangs. Users see an outage while the monitor stays green.

**4. Slow counts as up.** With a 30-second timeout, a page that takes 25 seconds to load counts as healthy. Nobody waits 25 seconds.

**5. Dependencies multiply.** If your app needs three services that are each at 99.9%, you’re at roughly 99.7% before your own code fails at all. That’s about **26 hours a year**, not 8 h 46 m.

**6. Maintenance windows “don’t count."** They don’t count against the SLA, but users still can’t use the service during them.

**What actually helped us:**

* 1-minute checks (or 30 seconds for anything customer-facing)
* Checking from 3+ regions and alerting only when 2+ agree
* Checking real endpoints (login, a key API call, a health check that touches the DB), not just `/`
* Alerting on response time, not just up/down
* Heartbeat monitoring for cron jobs and backups. A job that silently stops running is an outage nobody notices for weeks.
* Comparing the monitor’s uptime with the support ticket timeline after an incident. If they don’t match, the monitoring is wrong.99.9% uptime sounds great until you do the math: it allows 8h 46m of downtime a year, and your real number is probably lower
```



document fetched 2026-10-15 00:40 · canonical https://reddit.com/r/sysadmin/comments/1x5d60q/999_uptime_sounds_great_until_you_do_the_math_it
