What do uptime and SLA mean? The difference between 99.9% and 99.95%

Hosting pages say "99.9% uptime" and most people see it as marketing decoration. In fact there is a measurable commitment behind it — or there is not, and the two can be told apart.

What the percentages mean in minutes

Instead of an abstract percentage, look at the downtime allowed per year:

Uptime	Annual downtime	Monthly downtime
99.0%	3 days 15 hours	7 hours 18 min
99.5%	1 day 19 hours	3 hours 39 min
99.9%	8 hours 45 min	43 minutes
99.95%	4 hours 22 min	21 minutes
99.99%	52 minutes	4 minutes

The difference between 99.9% and 99.95% sounds small: 21 minutes a month instead of 43. On a game server, that is half of a busy evening.

The real question: a commitment, or a wish?

Whether an SLA is real is shown by one thing: what happens when it is not met? A percentage with no enforcement behind it is a wish, not a commitment.

Four items to look at:

Is there a service credit? If the commitment is not kept, how does it show up on the invoice?
How is it measured? Whose measurement is taken as the basis, and at what interval?
What is excluded? Planned maintenance, customer-caused problems, DDoS — it is normal for these to be out of scope, but it has to be in writing.
How is it claimed? Is the credit automatic, do you have to apply for it, is there a time limit?

A "99.9%" for which these four are not in writing cannot be verified.

Why is planned maintenance excluded?

Because not having it is worse. Security patches and hardware replacement require downtime at some point; the alternative is running on an unpatched system. What you should look at is not whether maintenance is excluded but whether it is announced in advance and how long it takes.

Measure it yourself

The provider's measurement and your own experience can differ; the provider measures its network, you measure your service. Setting up independent monitoring is not hard, and it becomes the only evidence you have in a dispute:

A free uptime monitoring service (an external check, at 1-5 minute intervals)
ping and a port check from a second server of your own
For a game server, a regular check against the query port

Do not run the monitoring from the server itself — when the server goes down, the monitoring goes down with it.

Where uptime is not a single number

Your server can be up while the service is unusable: the database is locked, the disk is full, the game process has crashed. An uptime commitment usually covers network and hardware reachability, not that your application is running.

That distinction marks out the area that is your responsibility: process monitoring, disk-full alerts and automatic restarts are on you. Adding Restart=always to a systemd service definition brings a crashed game server back within seconds and improves your uptime independently of the provider's commitment.

How this works at Datafex

The Service Level Commitment is built on 99.95% availability and a first response time of approximately 5 minutes; the measurement method, the out-of-scope situations and the service credit principles are written out item by item: Service Level Agreement (SLA)

The current status of the systems and planned maintenance announcements: system status