Guides

99.9% uptime: how much downtime is that, really?

Updated July 11, 2026 · 5 min read

Availability is stated in percent, and numbers close to 100 all look equally good. But the difference between 99% and 99.99% is the difference between three and a half days and 53 minutes of downtime per year. Here is the honest conversion.

Contents
  1. The conversion table
  2. Why the percentage misleads
  3. What is realistic?
  4. SLA percentages and measured uptime are different things
  5. Measuring uptime properly
  6. Frequently asked questions

The conversion table

This is how much downtime hides in the usual availability figures (based on an average month of 30.44 days):

AvailabilityDowntime per monthDowntime per year
99%approx. 7 h 18 minapprox. 3 days 16 h
99.5%approx. 3 h 39 minapprox. 1 day 20 h
99.9%approx. 44 minapprox. 8 h 46 min
99.95%approx. 22 minapprox. 4 h 23 min
99.99%approx. 4.4 minapprox. 53 min
99.999%approx. 26 secapprox. 5.3 min

Why the percentage misleads

99% sounds like an A grade, but it means: on average, the site is gone for more than seven hours every month. If that happens on a Saturday morning during the holiday season, the number on paper is still "99%".

Distribution matters as much as the total: one long outage is usually worse than many short ones, because it makes it into support calls, reviews and lost orders.

What is realistic?

Rough real-world reference points, no guarantees for any individual case:

  • Shared hosting: usually between 99.5% and 99.9%. Fine for many sites, but maintenance windows and noisy neighbours show up in the numbers.
  • A solid VPS or managed server: 99.9% and better is achievable if deployments are clean and someone notices outages quickly.
  • 99.99% and beyond: requires redundancy (multiple servers, load balancers, failover) and processes. That is an infrastructure project, not a hosting plan.

SLA percentages and measured uptime are different things

An SLA is a payment promise, not a measurement: if the provider misses the committed availability, you get credits, under their rules. Planned maintenance is often excluded, the measuring is done by the provider itself, and short incidents fall through the cracks.

Your own independent measurement does not replace the SLA, but it complements it: only your measurement shows what your users actually experienced, and only it gives you evidence for conversations with your host or your clients.

Measuring uptime properly

Two levers determine how solid your measured uptime is: the check interval and the confirmation logic. With 5-minute checks, a 4-minute outage can pass completely unnoticed; with 60-second checks it cannot. And counting every single failed request as downtime measures network noise, not availability, which is why good services confirm an outage with several checks before it enters the statistics.

Frequently asked questions

Is 100% uptime possible?

Over short periods yes, permanently no. Eventually there is a kernel update, a certificate rotation or an incident at your host. Credible providers make outages transparent instead of promising 100%.

What availability should I promise my clients?

Only one your infrastructure demonstrably delivers, with a margin. Promising 99.99% on a single server promises redundancy you do not have. An honestly measured 99.5% or 99.9% is worth more than 99.99% on paper.

Does planned maintenance count as downtime?

For your users: yes, the site is gone. In SLAs it is usually excluded. Good practice: announce maintenance (e.g. on your status page), schedule it at night and mark it as a maintenance window in your monitoring so the statistics stay honest.

Measure your real uptime.

Three monitors free forever, with 30-day statistics and exportable check logs.

Monitor for free