Blog · Operations

99.9% uptime: what it really means, and what really breaks

By the HostingAlly team Published 27 July 2026 7 min read

Three nines gets promised by every host going. Hardly any of them spell out what the figure means, what it leaves out, or which failures realistically hit a hosting platform. We run one, and the mystique around it serves marketing far better than it serves customers. So here is the operational view, put into English.

Do the arithmetic. 99.9% uptime leaves room for 43 minutes and 50 seconds of downtime per month. Move to 99.99% and you get 4 minutes 23 seconds. As for 100%, it has never existed. A host claiming it is telling you about an aspiration, a gap in their measurement, or a refund policy, and nothing about their infrastructure.

What actually takes websites down

Roughly in order of how often they happen on shared platforms. Application-level failures come first: a plugin update white-screens WordPress, so the site is 'down' while the server sits there perfectly healthy. Then resource exhaustion, where a neighbouring account or your own traffic spike runs into PHP process limits. Deployments and maintenance follow: kernel patches, PHP upgrades, and done well that means seconds of graceful restart. Hardware failures next, disks mostly, which is why RAID and real backups exist. Network events after that, upstream provider trouble and DDoS weather. Rarest of all, and most famous, the whole-datacentre incident.

That spread matters, because the marketing figure only covers part of it. What an uptime SLA measures is whether the server answered, not whether your site worked. A plugin that took out your checkout will never show up in a host's downtime ledger. Which is why daily backups and a staging environment buy you more effective uptime than the gap between 99.9% and 99.95% ever could.

How to read an uptime guarantee

Any SLA gives up its substance to three questions. First one: measured by whom? A guarantee scored on the host's own monitoring, with maintenance windows carved out, is a good deal softer than one you can verify independently. Put free uptime monitoring on your own site, ten minutes of work, and the ledger belongs to you.

Second question: what is the remedy? Most SLA credits amount to a pro-rata refund covering the downtime itself, which comes to pennies for an hour off the air. So the guarantee works as a statement of engineering intent, never as insurance. Judge a host on its status-page history and on how it talks to you mid-incident, not on the credit formula.

Third question: what gets excluded? Standard carve-outs cover scheduled maintenance, 'emergency' maintenance, DDoS, upstream failures and force majeure. The shorter that exclusion list runs, the more the headline number is worth.

What we actually do about it

Here is our own uptime posture, in concrete terms. NVMe RAID storage, so a single disk dying is a non-event. LiteSpeed serving cached pages, so most traffic spikes never touch PHP at all. Per-account resource isolation, so a neighbour's bad day stays their bad day. Continuous external monitoring, with alerting that wakes a human. And daily backups with self-service restores, which turns application-level failures, the most common kind by far, into ten-minute fixes instead of outages.

Now the honest limits. We run scheduled maintenance, announced in advance, seconds long, off-peak. Like everyone else, we lean on upstream networks we do not control. A determined enough DDoS counts as weather: mitigated, not prevented. Our commitment is 99.9% on every plan, because that is the figure we can stand behind across thousands of accounts, and we would sooner explain what it means than leave it sitting on the page as decoration.

Quick answers

99.9% uptime works out to how many minutes down?

Around 43.8 minutes a month, which comes to roughly 8 hours 46 minutes across a year. Push up to 99.99% and the allowance drops to 4.4 minutes a month. Engineering that extra step costs far more than the downtime itself costs most small sites. That is why 99.9%, paired with quick and honest recovery, remains the sweet spot on shared hosting.

Do uptime guarantees pay out meaningful compensation?

Rarely. A typical SLA credit is a pro-rata refund of the downtime period, and that comes to pennies. Read the SLA as an engineering promise you can hold a host to. Then verify it yourself with a free external monitor instead of trusting whatever the host's own dashboard reports.