What Is Uptime and a 99.9% SLA: What Your Host Actually Promises
"99.9% uptime" appears on almost every hosting website and sounds like "it basically never goes down." In reality that figure allows almost 9 hours of downtime a year, and what exactly counts as downtime — and what you get when the promise is broken — is written in the SLA. Here's how to read these promises.
What uptime and SLA mean
Uptime is the share of time a server or service was up and reachable. It's measured as a percentage over a period: a month or a year.
SLA (Service Level Agreement) is the document where the host states what uptime it guarantees, how it measures it and what it compensates if the promise isn't met.
The difference matters: an uptime figure on the homepage is marketing, an SLA is a contractual obligation. Without an SLA, the figure guarantees nothing.
How much downtime each figure allows
Allowed downtime per year and per month (30 days):
- 99% — 3 days 15 hours a year, 7 hours 12 minutes a month.
- 99.5% — 43 hours 48 minutes a year, 3 hours 36 minutes a month.
- 99.9% — 8 hours 46 minutes a year, 43 minutes a month.
- 99.95% — 4 hours 23 minutes a year, 22 minutes a month.
- 99.99% — 53 minutes a year, 4 minutes a month.
- 99.999% — 5 minutes a year, 26 seconds a month.
The gap between 99% and 99.9% looks like "just one more nine," but it's 3.5 days of downtime versus 9 hours. And a 99.999% promise from a regular VPS host isn't realistic: even a planned reboot takes longer than 26 seconds.
The fine print in an SLA
Two hosts can both advertise 99.9% and guarantee completely different things. What to check in the agreement:
What exactly is guaranteed. The data center network, the physical server, or your VPS itself? If the network is up but your server is frozen, under the first option there's no breach.
What doesn't count as downtime. Planned maintenance, DDoS attacks, upstream provider issues and customer-caused failures are almost always excluded. If maintenance isn't limited in length and frequency, the host can blame almost any outage on it.
How downtime is measured. By the host's monitoring or by your complaint? Sometimes downtime only counts from the moment you open a ticket, and short outages of a few minutes aren't counted at all.
Calculation period. 99.9% per month is stricter than 99.9% per year: in the second case the server can be down for almost 9 hours straight and still be within the guarantee.
What you get when the SLA is broken
Compensation is almost never cash — it's account credit, usually 5–30% of the monthly plan price depending on how long the outage lasted. You typically have to request it yourself within a limited window, or the right expires.
Keep in mind: an SLA doesn't cover your losses. If your store was down for a day and lost $500 in sales while the VPS costs $10, the compensation will be a few dollars. The real value of an SLA isn't the money — it's that the host is willing to commit to quality in writing.
How to check real uptime
Don't rely only on the host's numbers — keep your own record:
- External monitoring. UptimeRobot, Better Stack and similar services check your site every few minutes and alert you via Telegram or email when it's down. Their basic plans are free.
- The host's status page. Reliable providers have a public status page with incident history. If there's none, or it always says "all systems operational," that's a red flag.
- Server uptime since last boot. The command below shows how long the VPS has been running since its last start. An unexpectedly small value means the server was rebooted without your knowledge.
uptime -p
Boot and reboot history:
last reboot | head
Note the date, time and length of every outage: without that record, claiming SLA compensation is hard.
What uptime your project needs
- Blog, business-card site, test projects — 99.5–99.9% is enough.
- Online store, CRM, customer-facing bots — 99.9% or higher, with a proper SLA and fast support.
- Payment services, business SaaS — 99.95–99.99%. No host can deliver that on a single server: you need several servers in different data centers plus load balancing.
The cheapest insurance against downtime is regular backups to another server. With them, even a serious failure costs an hour or two of recovery.
FAQ
Is 100% uptime possible?
No. Every server sooner or later needs updates, hardware replacement or a reboot. A 100% promise is marketing, not a guarantee.
Does planned maintenance count as downtime?
With most hosts, no — as long as it was announced in advance. Check how many days' notice you get and how many hours per month maintenance can take.
What should I do if my host breaks the SLA?
Gather your monitoring data and contact support to request compensation without waiting for the end of the month. If outages keep happening, moving is cheaper than collecting credits.