- Turn every uptime percentage into minutes a month. 99.9% is 43 minutes; 99.99% is about 4.
- Response time is when someone replies. Resolution time is when it is fixed. Most agreements only commit to the first.
- RPO is how much data you could lose; RTO is how long you are down. Ask for both as numbers.
- Read the exclusions before the promises. Scheduled maintenance is usually left out of the uptime figure.
- A service credit refunds a fraction of your hosting fee, never your lost sales. Treat it as a signal, not compensation.
How do you read a hosting SLA?
A hosting SLA (service level agreement) is the provider's written promise about uptime, support and recovery. Read it in four passes: turn the uptime percentage into minutes, read the exclusions, separate support response from resolution, and get recovery targets as numbers. Most of what an SLA commits a provider to sits in the detail, not the headline.
What does an uptime percentage promise?
Very little until you turn it into time. 99.9% and 99.99% sound almost the same, but one allows about 43 minutes of downtime a month and the other about 4. Convert every figure you are quoted, because that is the only form you can compare.
| Uptime | Per month | Per year | In practice |
|---|---|---|---|
| 99% | 7h 18m | 3d 15h | A working day of downtime a year |
| 99.5% | 3h 39m | 1d 20h | Noticeable to regular visitors |
| 99.9% | 43m | 8h 46m | The common shared-hosting figure |
| 99.95% | 22m | 4h 23m | Better managed hosting |
| 99.99% | 4m 23s | 52m | High availability |
| 99.999% | 26s | 5m 15s | Rare outside large infrastructure providers |
The second decimal place is where the number starts to mean something. 99% is a very low bar.
Ask what "up" means, and who measures it
Some agreements count a server as up if it answers a ping, which it will do while your site shows an error to every visitor. Others measure a full page load. A check run from inside the provider's own network never follows the path your visitors use. Ask what is checked, from where, and how often.
How it is measured matters as much as the figure. A check every five minutes can miss a four-minute outage, and a check from one place cannot tell you whether visitors elsewhere could reach you. G7Cloud commits to 99.99% uptime and checks every site every minute, with a public status page. Our SLA sets out the terms.
What is excluded from the uptime figure?
Read this part before the headline percentage. A generous figure with broad exclusions can commit a provider to less than a modest figure with narrow ones.
| Common exclusion | Reasonable? | What to ask |
|---|---|---|
| Scheduled maintenance | Yes, with limits | How much notice, and is there a monthly cap? |
| Your own code or plugins | Yes | Who decides, and what evidence do they give? |
| DDoS and other attacks | Sometimes | Is protection included, or is an attack excluded as force majeure? |
| Third-party network failures | Partly | Their network provider is still their supplier |
| DNS not under their control | Yes | Fair if you host DNS elsewhere |
| Anything caused by "emergency maintenance" | No | Undefined and unlimited. Ask for a definition |
The last row is the one to push back on. Without a definition it can cover any outage after the fact.
- How much notice for planned maintenance? A visitor who cannot check out at 2am does not care that the window was announced. Seven days is reasonable. Twenty-four hours is not, for anything that takes the site offline.
- Is there a monthly cap on excluded maintenance? Without one, the exclusion has no limit.
- Who declares an outage? If only the provider can, and you must claim within a short window, the burden is on you.
- Is there a public status page with past incidents? A provider that publishes its history expects to be held to it.
How do I read support response times?
Separate response from resolution. Response time is how long until a person replies. Resolution time is how long until the problem is fixed. Almost every agreement commits to the first and says nothing about the second, which is worth knowing before you rely on it.
| Term | Means | Watch for |
|---|---|---|
| First response time | Someone acknowledges the ticket | An automatic reply that stops the clock |
| Resolution time | The problem is fixed | Rarely committed to. Ask if it is |
| Business hours | Their office hours, in their time zone | Which time zone, and which public holidays |
| 24/7 support | Someone is reachable at any hour | Reachable by whom, and can they act or only pass it on? |
| Priority levels | How the ticket is classified | Who classifies it, you or them? |
Rows four and five decide whether 24/7 support means an engineer or an inbox.
- Ask who answers out of hours. An engineer who can restart a service is a different thing from someone who logs the ticket for the morning.
- Ask who sets the priority. If the provider classes your total outage as normal priority, the emergency response time never applies.
- Ask what happens on a UK bank holiday. "Business hours" over a long weekend can mean four days.
- Ask for the escalation path in writing. What matters is what happens when the first reply does not fix it.
Test it before you need it
Send a fairly technical question before you buy, and see how long a useful answer takes and whether it comes from someone who knows the platform. That one test tells you more about an incident than any published response time.
What do RPO and RTO mean for backups?
Two numbers describe any recovery promise. RPO (recovery point objective) is how much data you could lose: the gap between the last good backup and the failure. RTO (recovery time objective) is how long you are down while it is restored. Ask for both as numbers. "Regular backups" commits a provider to nothing.
| Backup frequency | Worst-case data loss (RPO) | Suits |
|---|---|---|
| Weekly | 7 days | Nothing you would miss |
| Daily | 24 hours | Brochure sites, content that rarely changes |
| Every 6 hours | 6 hours | Active content sites |
| Hourly | 1 hour | Busy sites and small shops |
| Continuous replication | Seconds | Systems that take payments or bookings all day |
For a shop, ask what a day of lost orders is worth, then compare it with the cost of more frequent backups.
- 1How often are backups taken, and how long are they kept? Daily, kept for thirty days, is a common and reasonable answer.
- 2Are they held somewhere separate? A backup on the same server, or in the same data centre, does not survive the events that matter most.
- 3Are they tested? A backup can run for months and still fail to restore: a corrupt archive, a missing database, half the files. Ask whether restores are checked automatically, how often, and what happens when a test fails.
- 4How long does a restore take, and who runs it? Restoring it yourself in minutes and waiting hours on a support ticket are very different services.
- 5Can you restore one file or one table? If you can only restore the whole site, a small mistake costs you every change since the backup.
On G7Cloud every site gets daily backups you restore yourself from the dashboard, down to a single file, folder or database. Every night we check backups for corruption and restore a rotating sample into a throwaway test copy, so a broken backup is found before you need it. See how our backups work and our backup strategy guide.
Are service credits worth anything?
As compensation, no. A service credit usually refunds a percentage of one month's fee, which on a small plan is a few pounds. If four hours of downtime costs a shop several thousand pounds in orders, the credit does not come close. Its value is as a signal: a provider willing to put money behind a number expects to hit it.
| Question | Why it matters |
|---|---|
| Is the credit automatic or claimed? | With a claims process and a short window, most credits are never paid |
| What is the claim deadline? | Some require notice within days of the incident |
| What is the maximum? | Usually capped at one month's fee, whatever the impact |
| Credit or cash? | A credit only helps if you are staying |
| Does accepting it waive other remedies? | Some agreements make the credit your only remedy |
Read the last row twice on any large contract.
How we handle this
On self-serve plans we commit to 99.99% uptime, backed by checks every minute and a 30-day money-back guarantee, rather than a service-credit table that would return a few pounds. Enterprise contracts can carry a written SLA with agreed service credits, set per contract. G7Cloud is hosted in the UK, on servers we own and run ourselves. See our SLA or Enterprise.
What should I check before signing a hosting SLA?
- 1Turn every percentage into minutes a month, so you compare like with like.
- 2Read the exclusions first, then the headline figure in that light.
- 3Separate response from resolution, and find out who answers out of hours.
- 4Get RPO and RTO as numbers, and ask whether restores are tested.
- 5Check for a public status page with incident history. It is the cheapest evidence of how a provider behaves in a bad week.
- 6Test support before you commit, with a question that needs a proper answer.
The other half of reliability is what you can see yourself. A provider's status page tells you when they know something is wrong. Your own monitoring tells you when your site is wrong for reasons they will never see. Checking CPU, memory and disk usage and reading system logs cover the server side.
Frequently asked questions
How much downtime does 99.9% uptime allow?
About 43 minutes a month, or 8 hours 46 minutes a year. 99.99% allows about 4 minutes 23 seconds a month. Always turn a percentage into minutes before comparing providers, because the decimal places hide a tenfold difference.
What is the difference between response time and resolution time?
Response time is how long until someone replies to your ticket. Resolution time is how long until the problem is fixed. Nearly every agreement commits only to response, so a one-hour response time says nothing about how long you will be down.
What do RPO and RTO mean?
RPO (recovery point objective) is the most data you could lose, which depends on how often backups run. RTO (recovery time objective) is how long recovery takes. Daily backups mean an RPO of 24 hours: everything since the last one is gone.
Are hosting service credits worth claiming?
Financially, rarely. They refund a fraction of one month's fee, capped well below the cost to your business, and many must be claimed within days. Their value is as a signal of confidence. Check whether accepting one waives your other remedies, as some contracts say it does.
Is scheduled maintenance counted as downtime?
Usually not: most agreements leave announced maintenance out of the uptime figure. Ask how much notice you get, whether there is a monthly cap on those hours, and how "emergency maintenance" is defined, since an undefined term can cover any outage after the fact.
About G7Cloud Engineering
Articles written by the engineers who build and run G7Cloud: UK managed hosting and the AI Website Builder. We write about what we operate every day: containers, backups, databases, and the small-business websites that run on them.
More about G7Cloud →