Behind Our 99.9% Uptime Promise: What It Means and How We Keep It

Every host prints 99.9%. Here is what stands behind ours.

We have put 99.9% uptime on our plans since 2008, and so has almost everyone else, so the number on its own tells you very little.

The difference is what a host built to hit it, whether it measures it where you can see, and what it does in the first fifteen minutes when something breaks. We can show you all three, which is more than most hosts can do for any one of them.

First, the arithmetic

99.9% is not “always”. A month has 43,800 minutes. One tenth of one percent of that is 43.8 minutes. That is the downtime a 99.9% promise allows, every month, with the promise intact.

Promise	Allowed downtime per month	Per year
99%	7.3 hours	3.65 days
99.5%	3.65 hours	1.83 days
99.9%	43.8 minutes	8.76 hours
99.99%	4.4 minutes	52.6 minutes

The step from 99% to 99.9% is where the money goes: redundant power, redundant transit, hardware that survives a failed part, people awake at 3am. We chose 99.9% because it is the number we can build for and back with our own money. We do not print 99.99% on a shared plan, and we don’t think anyone should.

What we built to stay under 43.8 minutes

Uptime comes less from policy than from a long run of decisions about where servers sit and what they are plugged into. These are the ones we made.

Facilities. More than 10 datacenters, each with its own certification (ISO 27001 or Tier III, listed per location on our datacenters page) and N+1 redundant power and cooling. Every critical component has a running spare, so a failed feed or chiller is absorbed, not felt. The certification and redundancy for each site are on the datacenters page.

Network. Tier-1 transit at every location from carriers such as NTT, GTT, Cogent and DE-CIX, with DDoS protection in front. One carrier having a bad afternoon does not take a site offline. The network page lists every location with a speed test you can run against it right now.

Hardware. We replaced every spinning drive on the platform with NVMe years ago. That gets sold as speed, and it is, but it is also reliability: no moving parts, fewer failures, faster rebuilds when a drive does go. VPS and Cloud customers get dedicated KVM resources, so a noisy neighbour cannot eat your CPU and turn a busy hour into an outage.

Geography. North America, Europe and Asia Pacific. A problem in one region stays in one region, and customers who need more than one location can run in two. We help set that up.

The same hardware produces the numbers in how fast GOZEN hosting actually is. Speed and uptime come from the same decisions.

How we measure it, where you can see it

Most hosts skip this part, which is a shame, because it matters more than anything above.

A host that measures uptime from inside its own network never sees a failed transit link, a broken resolver, or a datacenter cut off from the internet. The number looks wonderful and describes nothing you experienced. So we measure from outside.

Our status page at status.gozenhost.com is fed by monitors that check each service from beyond our network. It lists 28 services by name: shared hosting in the US and EU, the VPS nodes, every cloud node, IMAP, SMTP and webmail, six DNS servers, the client portal, the Git server and our own network tools. When something goes down, it goes down there first, in front of everyone.

The page keeps 120 days of history for every service and states the overall figure in one line at the top. As we write this, on 12 September 2026, that line reads: delivered 99.98% over the last 90 days. By group, over the last 120 days: DNS 100.00%, cPanel hosting 99.99%, email 99.98%, core services 99.97%, cloud hosting 99.96%. Every incident behind those numbers is listed underneath with its date and how long it lasted, most of them a few minutes. Go and look. The figures will have moved by the time you read this, and they are supposed to.

Our SLA names that page as the record. There is no private figure that differs from the public one.

The SLA also says exactly what counts. Downtime is five or more consecutive minutes with no packet delivery to your service. A single dropped check during a router failover, which nobody actually experienced, does not count, and neither does a vague “service degradation.” Five sustained minutes is something your visitors would notice, so that is where our clock starts, and it is written down so there is nothing to argue about afterwards.

Maintenance we announce in advance is excluded, which is standard. What you will not find in our terms is an exclusion for “third-party providers” or “conditions outside our control” worded broadly enough to exclude the internet itself.

The first fifteen minutes

The uptime percentage describes last month. What decides how your next outage goes is how quickly a person with root access is looking at it.

We commit to a human response on a production-down ticket within 15 minutes, 24 hours a day, in writing, on the SLA page. It is usually faster, because the same monitors that feed the status page page us before the ticket arrives. Managed VPS and Cloud customers have 24/7 server administration included, so the person who answers can act on the box, not just reply.

“We aim to respond as quickly as possible”, which is what a lot of terms pages say, is not a commitment so much as a mood.

If we miss, this is what you get

The schedule is public and gets checked against the status page record.

Monthly uptime	Downtime	Credit
99.9% or better	up to 43.8 minutes	none, SLA met
99.5% to 99.89%	up to 3.6 hours	10% of the month’s fee
99.0% to 99.49%	up to 7.2 hours	25%
95.0% to 98.99%	up to 36 hours	50%
below 95%	more than 36 hours	100%

Claims go through a support ticket under Billing, SLA Credit Claim.

Be realistic about what a credit is worth: on a $20 plan, three hours down returns $2. That does not make up for a lost afternoon of sales, and no host’s credit would. It is our money on the line for a number we publish, which is how you know we measure it and expect to be held to it. If your business cannot afford an hour offline, ask us about running in two locations. It protects you far better than any credit schedule, and we would rather set it up than pay out afterwards.

Check us

You do not need to take our word for any of this.

Point an external monitor at your site. The free tiers of UptimeRobot and Better Stack are enough, and if you would rather own the data, Uptime Kuma is open source and runs on the smallest VPS we sell. Our knowledge base walks through monitoring your own server, Uptime Kuma included. Check a real page rather than a ping, every one to five minutes from two locations, and read the number after 30 days.

Then open our status page and read the history, not just the colour. A page that has only ever been green, with no incidents written up, tells you nobody writes incidents up. Ours shows what happened, when, and for how long.

We hold our backups to the same standard. The backup version of this post is just as blunt, and includes a ten-minute test.

The short version

99.9% allows 43.8 minutes a month. We built more than 10 redundant locations on Tier-1 transit and NVMe to stay well under it. We measure it from outside, on a public page our SLA is tied to, and that page read 99.98% over the last 90 days when this was written. We answer a production-down ticket within 15 minutes. And we publish exactly what you get back if we ever miss.

If your current host prints the number but cannot show you any of that, we will move your site for free, with no downtime, and you can watch our monitors while we do it.