Almost every hosting company in the country promises 99.9% uptime. Written down like that it reads as a promise never to be offline.
Do the sums and it is a promise to be offline for up to eight and three quarter hours a year, at a moment nobody consults you about, with compensation measured in pennies if you spot it and put in a claim.
That is not an accusation aimed at anyone in particular. Ours are real servers too and real servers have bad days. The problem is that the figure has stopped carrying information, because everybody quotes it and hardly anybody defines it. So here is the arithmetic, then some questions that get you a far more useful answer.
The arithmetic, shown properly
A year is 365 days. Times 24, that is 8,760 hours. One tenth of one per cent of 8,760 hours is 8.76 hours, which is eight hours and about forty-six minutes.
Run the same calculation across the figures you see quoted and the ladder looks like this:
- 99% allows 87.6 hours a year. Three and a half days.
- 99.5% allows 43.8 hours. Nearly two full days.
- 99.9% allows 8.76 hours. A working day, give or take.
- 99.99% allows 52.6 minutes.
- 99.999% allows about five and a quarter minutes, and costs a fortune to engineer.
Each extra nine divides the allowance by ten, which is why the jump from three nines to four is such a big deal in the machine room and such a small deal on the sales page.
Most contracts measure by the month rather than the year, and that flatters the number nicely. A 30-day month contains 720 hours, so 99.9% of it permits 43 minutes off. Lose 43 minutes in March, another 43 in April, and the guarantee has been met both times while you have quietly lost an hour and a half of trading.
What the host counts as being down
This is where the small print does its work, because the percentage only applies to whatever the contract has defined as an outage. Typical carve-outs include planned maintenance windows, anything caused by your own code or plugins, network problems belonging to somebody else, attacks, and anything to do with your domain or DNS.
Strip all that out and the guarantee often covers a narrower slice of reality than you imagined.
Measurement matters just as much. “Up” usually means the server answered a request from a monitoring station. Your site can answer that request while the database has fallen over and every page returns an error. It can answer while the shopping basket is broken. It can answer after thirty-two seconds, by which point the visitor has gone back to Google. All three of those count as up, and all three mean you are shut.
Email is the one that catches people out. A great many uptime clauses cover the web server only, so mail can stop flowing for a morning without denting the figure at all.
The refund is not worth chasing
Service level agreements pay out in account credit, worked out against what you pay them. Take a budget plan at £2.99 a month. Divide that by the 720 hours in the month and you are paying roughly four tenths of a penny an hour. Four hours offline earns you a credit of about 1.7p.
Some hosts multiply it. Ten times almost nothing is still almost nothing.
The same maths applies to us, by the way. At £1 a day, four hours of downtime is worth about 17p of your money, and we would not insult you by posting a cheque for it. Nobody is made whole by an SLA credit. The cost of a morning offline has nothing whatsoever to do with the size of your hosting bill, which is precisely why the credit is the least interesting clause in the contract.
Worth knowing too: credits are rarely automatic. You generally have to notice the outage yourself, keep some evidence, and submit a claim inside whatever window the terms specify. Most people never bother, which is presumably part of the appeal.
Five questions that tell you more than a percentage
Ask these instead. The answers separate hosts far more reliably than any number on a pricing table.
What exactly are you checking, and how often?
Pinging a server proves the machine has power. Loading a real page, checking the words on it are the words that should be there, and confirming mail is still moving proves the business is open. Ask which of those they do, and how many minutes pass between checks.
Where does the alarm go at three in the morning?
Round-the-clock monitoring can quite legitimately describe a graph on a screen that nobody looks at until nine. The question is whether the alert wakes a human being, and which human being it wakes. Ours is genuinely watched around the clock, and outside office hours the alert reaches whoever would have dealt with it in the middle of the working day.
Is that person allowed to fix it?
Plenty of first-line staff can only log a ticket and escalate. If the person who answers cannot get onto the server, the clock keeps running while your problem travels up a chain.
What happens when the fault is mine?
Sites break because of a plugin update far more often than because a data centre catches fire. A budget host will tell you the server is fine and close the ticket, which is technically true and completely useless. A managed host treats a broken site as a broken site. We look after WordPress hosting on that basis, because a customer who cannot buy anything does not care whose fault it is.
How many other sites share my server?
Nobody publishes this and most hosts will dodge it. A vague answer is still informative. Packing thousands of sites onto one box is how the ultra-cheap price gets built, and it is also how your Tuesday gets ruined by somebody else’s traffic spike.
What we say instead of a number
We have run our own servers for over twenty years and we keep hundreds of sites going every day. Things still go wrong occasionally, because that is what hardware and software do.
So rather than paint a percentage on the wall, we would rather tell you what happens on a bad day: somebody notices before you do, that somebody can log straight into the machine, and you can ring 01623 650 333 and get a person who already knows your setup. That is a harder thing to put in a comparison table and a much better thing to own.
Uptime Guarantee FAQs
How much downtime does a 99.9% uptime guarantee allow?
A year contains 8,760 hours, and one tenth of one per cent of that is 8.76 hours. So a 99.9% guarantee permits roughly eight hours and forty-five minutes offline per year without the host breaking its promise. Measured monthly, the same figure works out at about 43 minutes in a 30-day month.
Is a 99.9% uptime guarantee good?
It is an ordinary target rather than an impressive one, because nearly every host quotes it. What matters more is the definition underneath: what the host counts as downtime, what it excludes, how it measures, and what it does when your site breaks for reasons that are not the server. Those answers vary enormously.
What do I get if my host misses its uptime guarantee?
Usually account credit worked out from your monthly fee. On a cheap plan, several hours offline can be worth a few pence. Credits are rarely paid automatically either, so you normally have to notice the outage, gather evidence and claim within a set window. Treat the guarantee as a statement of intent rather than insurance.
Does uptime monitoring include email?
Often not. Many uptime clauses cover the web server alone, so mail can stop for hours without affecting the published figure. If email matters to your business, ask specifically whether it is monitored, what is checked, and who is alerted when delivery stops. It is a fair question and a good host will answer it plainly.
Ask us the five questions
Put them to your current host first. If the replies leave you uneasy, put the same five to us and compare. Monitoring is part of the managed service, which begins at £1 a day + VAT, and we will happily walk you through what gets watched and who gets woken up.
Leave a Reply