
Almost every hosting plan advertises an uptime figure. It is usually 99.9%, sometimes 99.95% or 99.99%, and it sits on the pricing page next to the storage and bandwidth numbers as if it were just as easy to verify. It is not. Your host measures its own uptime, decides what counts as an outage and decides which hours were scheduled maintenance. If you want to know whether the promise is being kept, and whether you are owed a credit, you need numbers of your own.
What the Percentages Actually Allow
Uptime guarantees sound almost identical until you convert them into minutes. Over an average month of about 30.4 days:
- 99.9% allows roughly 43 minutes of downtime.
- 99.95% allows roughly 22 minutes.
- 99.99% allows a little over 4 minutes.
So a “three nines” host can be down for most of an hour every month and still be inside its promise. A single bad evening, with the site failing on and off for 90 minutes, breaks it twice over. Knowing which side of the line you are on is the whole point of measuring.
Read the Fine Print Before You Measure
An uptime guarantee is part of the service agreement, and the agreement decides what you can claim. Before you set anything up, look for four details:
- What is covered. Many guarantees cover the network or the physical server, not your website. If PHP crashes or the database runs out of connections, the host may argue that the server was up.
- Maintenance windows. Scheduled maintenance is usually excluded, and some hosts only need to announce it a few days ahead.
- How downtime is measured. Some agreements only count outages longer than a few minutes, or only outages confirmed by the host’s own monitoring.
- How to claim. Credits are rarely automatic. You typically have to open a ticket within a fixed period, often 30 days, and provide dates and times.
That last point is where most customers lose out. Without your own log of when the site was unreachable, a credit request becomes a disagreement you cannot win.
Measure From Outside the Host’s Network
A check that runs on the same server, or inside the same data centre, shares the failure it is supposed to detect. If the server loses its upstream connection, a script on that server will happily report that everything is fine. Useful uptime data comes from a machine on a different network that requests your site the way a visitor would.
A good external check does more than ping the IP address. It should:
- Request a real page over HTTPS and check the status code, so a 500 error or a hosting “account suspended” page counts as down.
- Validate the TLS certificate, because an expired or incomplete certificate chain stops browsers and API clients just as effectively as a crash.
- Follow redirects to the final page instead of stopping at the first 301.
- Record the response time, since a site that takes 25 seconds to answer is unusable even if it technically responds.
Optionally, add a keyword check on an important page. Some failures return a perfectly normal 200 status with an error message in the body, and only a content check catches those.
Pick an Interval That Matches the Promise
The check interval puts a limit on how precise your numbers can be. With checks every five minutes, a two-minute outage can fall between two checks and never appear in your log, and the outages you do see are only accurate to within a few minutes. That is fine for confirming a 99.9% guarantee, where the monthly allowance is over 40 minutes. For a 99.99% guarantee, with an allowance of about four minutes, you need checks every minute or faster to say anything meaningful.
Filter Out False Alarms
The internet between your monitor and your host is not perfectly reliable. A single failed request can be caused by a dropped packet on a route that has nothing to do with your hosting provider. If you open a ticket for every blip, the support team will soon stop taking your reports seriously.
The usual fix is confirmation: when a check fails, the monitor repeats it a short time later and only records an outage if the second attempt fails too. This removes most one-off network noise while still catching real incidents within a minute or so. It is also worth remembering where your checks come from. A monitor located on another continent will report higher response times than your local visitors see, simply because of distance, so compare response times with each other over time rather than with a number from the host’s marketing page.
Keep a Record You Can Use
When you do find an outage, the evidence has to be specific. For each incident, note:
- When it started and ended, with the time zone.
- What the check saw: a timeout, a connection refused, a 502 or 503 status, or a certificate error.
- The total downtime for the month, compared with the allowance in your agreement.
A monitoring service that keeps this history and sends a recovery alert with the total duration does most of the work for you. A public status page is a useful extra: your own visitors can check it instead of emailing you, and it gives you a timestamped record that is easy to link in a support ticket.
Setting It Up Takes Minutes
You do not need your own infrastructure to do any of this. Hosted monitoring services handle the scheduling, the confirmation checks, the alerts and the history. FastPing, for example, checks a website’s status code, certificate and response time, repeats any failed check 20 seconds later before it opens an incident, and keeps a status page and incident history. Its free plan covers three monitors at five-minute intervals, which is enough to watch a small site, its certificate and its domain expiry date, with faster intervals on paid plans. Whatever tool you choose, the setup is similar: add the URL, choose the interval and the alert channel, and let it collect data for a full billing month before you draw conclusions.
What to Do With the Results
After a month or two you will have one of three outcomes. Your host may be comfortably inside its guarantee, in which case you have peace of mind and an alerting system for the day something does break. Your host may be over the limit, in which case you can file a credit request with dates, durations and error types instead of a vague complaint. Or you may find that the outages are short and frequent rather than long, a pattern that often points to an overloaded shared server, and a good reason to ask about a move to a different node or plan before the problem grows.
Either way, the uptime figure on the pricing page stops being a promise you have to trust and becomes a number you can check.
