Skip to main content

Blog · Operations

99.9% uptime: what that figure permits, and what actually fails

By the Hosting & Domains team Published 27 July 2026 7 min read

Three nines appears on the homepage of every host. Hardly any of them explain what the figure covers, what falls outside it, or how a hosting platform fails in real life. We run a platform ourselves. The fog around that number does more for marketing than for the people paying, so what follows is the view from operations, said plainly.

Start with the arithmetic. 99.9% uptime permits 43 minutes and 50 seconds of downtime a month. Move to 99.99% and the allowance drops to 4 minutes and 23 seconds. As for 100%, no such thing has ever existed. Any host advertising it is telling you about an ambition, a blind spot in its monitoring, or a refund policy, and not about the platform.

What actually knocks websites offline

What follows is ordered by how often we see each one on shared platforms. Application faults lead: a plugin update white-screens WordPress, the site goes dark, and the server underneath carries on without a care. Resource exhaustion comes next, when a surge of your own traffic or a neighbouring account collides with PHP process limits. After that, planned work. Kernel patches and PHP upgrades, which handled well cost you seconds of graceful restart. Then hardware, disks more often than anything else, which is the reason RAID and proper backups exist. Then network events, meaning upstream carrier trouble and DDoS. Whole-datacentre incidents come last, and they are the ones everybody has heard of.

That order matters, because the advertised figure only reaches part of the list. An uptime SLA records one thing: did the server answer? Whether the site worked is a separate question, and no downtime column anywhere has ever logged the plugin that broke your checkout. So daily backups and a staging environment will do more for the uptime your readers feel than the distance between 99.9% and 99.95% ever manages.

Reading an uptime guarantee properly

Any SLA comes down to three questions. The first: who keeps score? A guarantee measured by the host's own monitoring, with maintenance windows cut out of the total, carries less weight than one you can audit for yourself. Spend ten minutes putting free uptime monitoring on your site and the scorecard is yours.

The second: what follows when the guarantee is missed? A standard credit refunds the downtime pro rata, so an hour offline returns pennies to you. Treat that pledge as a statement of engineering intent, not as insurance. What deserves your judgement is the status-page record and the way a host speaks to customers mid-incident. The credit formula tells you very little.

The third: what has been carved out of the total? Scheduled maintenance, emergency maintenance, DDoS, upstream failures, force majeure. The shorter that list runs, the more the headline number is worth.

How we run our end of it

Here is the arrangement in concrete terms. NVMe RAID storage, so a dead disk changes nothing for you. LiteSpeed serving pages from cache, which keeps most traffic spikes away from PHP altogether. Resource isolation per account, so a neighbour having a hard afternoon does not drag you into it. External monitoring runs without a break, wired to alerts that wake a real person up. Nightly backups with restores you run yourself, which turn the most common failure class of the lot, the application kind, from an outage into a ten-minute rollback.

Now the limits, stated plainly. Scheduled maintenance does happen here, announced ahead of time, run off-peak, and measured in seconds. Upstream networks carry our traffic exactly as they carry every other host's. A determined DDoS is weather: we absorb it and blunt it, and nobody makes it go away. Our commitment is 99.9% on every plan, because that is the level we can hold across thousands of accounts, and explaining the number beats leaving it on the page as decoration.

Quick answers

How much downtime does 99.9% uptime allow?

Around 43.8 minutes a month, which works out at roughly 8 hours 46 minutes across a year. At 99.99% the allowance is 4.4 minutes a month. Engineering that gap away costs a great deal more than the downtime costs a small site, so 99.9% paired with fast, honest recovery is the sensible target on shared hosting.

Is an uptime guarantee's payout worth anything?

Rarely. A typical credit hands back the downtime pro rata, and that amounts to pennies. Treat the SLA as an engineering promise you can hold a host to, then measure the result yourself with a free external monitor instead of trusting the provider's own dashboard.

Up next

More from the blog

The hosting these notes are written on

Renewals that stay flat, limits printed before you buy, migration at no charge and a support desk that writes back, all wrapped into one plan.

Browse Hosting Plans