Web hosting: how to choose a reliable solution for an SME?
Compare web hosting based on actual service: maintenance, backups, support, and site recovery, rather than just price or server power.
Sur cette page6 sections
When comparing two hosting plans, it's easy to see the monthly price, disk space, and a few reassuring words about security. That's handy for filling out a spreadsheet. It's much less helpful for answering the question that will actually matter on the day of a crash: who gets my site back up and running, and with what?
For a small business, the right hosting isn't necessarily the most powerful. It's the one that matches how the site is used and what you want to handle yourself. Renting a server and asking someone to look after your site are not the same service, even when both appear on the same invoice.
A host can provide a perfectly operational server while your contact form stops working after an update. Conversely, a developer cannot fix a host's network outage within the code. The two areas overlap, but responsibilities must remain clear.
This is why "hosting included" requires a bit of explanation. Included with what? Backups? Domain renewal? Site updates? Support if enquiries stop coming through? The same label can cover very different services.
A virtual server (VPS) gives more control over an isolated environment, while a dedicated server reserves a physical machine. These solutions can be relevant for certain traffic volumes or technical needs. They also bring things to manage: the system, updates, monitoring, and configuration, depending on what the provider handles.
Managed services (or "infogérance") describes this operational work entrusted to a provider. It's not an extra type of machine. You can have a modest infrastructure that is well-monitored, or a very powerful server where no one checks the alerts.
I prefer a solution where the limits are known and managed over a big machine chosen for its name. Power matters when you need it. It won't answer support emails for you.
Monitoring must therefore be adapted to what the site does. A showcase site, a booking system, and an online shop don't have the same functions to monitor, nor the same consequences when they stop.
Ask which channel to use, at what times, and what is planned depending on the severity of the incident. The initial response time is not the resolution time. "We have received your request" is not yet "your payment system is working again".
We cannot guarantee that no problems will occur. We can, however, specify who receives alerts, who works on the server, who checks the application, and how you are kept informed. This organisation is worth more than an adjective like "premium" on a sales page.
Backup frequency must therefore match what you can afford to lose. A presentation page that rarely changes and an application receiving orders all day pose different questions. You also need to check what is being copied: files, photos, database, and the elements needed to rebuild the whole thing.
A copy kept only on the server that crashes is not enough. And a backup that no one has tried to restore leaves a lot of uncertainty: is it complete, readable, recent enough? The CNIL notably recommends protecting copies and testing their restoration.
Ask for a concrete example of a restoration, including the scope involved and the checks performed. We aren't looking for a spectacular promise, but proof that a procedure exists and that it actually works.
You must distinguish between a data export and reusable software. On a closed service, you might be able to download your content without retrieving the site's functionality. This isn't necessarily a bad choice, but you need to understand it before building your entire organisation around it.
The location of servers and backups also deserves a precise answer, especially when the site processes customer information. "In Europe" alone does not describe access, the providers involved, or the applicable commitments. Ask for the documents and the scope of the service rather than assuming compliance from a map with a flag.
Then add normal operations: renewing the domain, applying patches, increasing resources, transferring the site. You will quickly see if the displayed price matches your needs or simply the rental of a piece of infrastructure.
At Cadarsir, I prefer to define these responsibilities right from the start of the project rather than letting you discover them during an incident. Truly included monitoring must remain explicit; not all sites require the same setup.
And if your problem is primarily a slow site, start by looking at what the browser experiences. A better server can help. It won't make a page light if it sends too many images and scripts. Choosing hosting is about finding a suitable base and people capable of operating it, not looking for a machine that will fix everything on its own.
For a small business, the right hosting isn't necessarily the most powerful. It's the one that matches how the site is used and what you want to handle yourself. Renting a server and asking someone to look after your site are not the same service, even when both appear on the same invoice.
The address, the premises, and the maintenance
The domain name is your site's address. Hosting provides the resources that allow it to run. Maintenance concerns the software: its updates, fixes, and monitoring. You can compare it to a shop: having an address and renting the premises doesn't mean the landlord also looks after your till and your stock.A host can provide a perfectly operational server while your contact form stops working after an update. Conversely, a developer cannot fix a host's network outage within the code. The two areas overlap, but responsibilities must remain clear.
This is why "hosting included" requires a bit of explanation. Included with what? Backups? Domain renewal? Site updates? Support if enquiries stop coming through? The same label can cover very different services.
Shared or dedicated server: it's not a contest
With shared hosting, several sites share an infrastructure and a portion of its resources. This can be perfectly suitable for a showcase website. You don't need an entire building to present your services and receive a few enquiries.A virtual server (VPS) gives more control over an isolated environment, while a dedicated server reserves a physical machine. These solutions can be relevant for certain traffic volumes or technical needs. They also bring things to manage: the system, updates, monitoring, and configuration, depending on what the provider handles.
Managed services (or "infogérance") describes this operational work entrusted to a provider. It's not an extra type of machine. You can have a modest infrastructure that is well-monitored, or a very powerful server where no one checks the alerts.
I prefer a solution where the limits are known and managed over a big machine chosen for its name. Power matters when you need it. It won't answer support emails for you.
During a breakdown, everyone must know what to do
Take a shop where the payment system fails while the product pages remain accessible. A check that only opens the homepage might conclude that everything is fine. The visitor, however, can no longer buy.Monitoring must therefore be adapted to what the site does. A showcase site, a booking system, and an online shop don't have the same functions to monitor, nor the same consequences when they stop.
Ask which channel to use, at what times, and what is planned depending on the severity of the incident. The initial response time is not the resolution time. "We have received your request" is not yet "your payment system is working again".
We cannot guarantee that no problems will occur. We can, however, specify who receives alerts, who works on the server, who checks the application, and how you are kept informed. This organisation is worth more than an adjective like "premium" on a sales page.
Backups must allow for a fresh start
Imagine your site comes back online after an incident, but without the week's bookings. Technically, it works. For your business, a piece of work is missing.Backup frequency must therefore match what you can afford to lose. A presentation page that rarely changes and an application receiving orders all day pose different questions. You also need to check what is being copied: files, photos, database, and the elements needed to rebuild the whole thing.
A copy kept only on the server that crashes is not enough. And a backup that no one has tried to restore leaves a lot of uncertainty: is it complete, readable, recent enough? The CNIL notably recommends protecting copies and testing their restoration.
Ask for a concrete example of a restoration, including the scope involved and the checks performed. We aren't looking for a spectacular promise, but proof that a procedure exists and that it actually works.
You must be able to move elsewhere too
An offer is easier to choose when you know how you could leave it. Does the domain really belong to the company? Who has access to its management account? Can the code, media, and data be retrieved in a usable format? How much would a migration cost?You must distinguish between a data export and reusable software. On a closed service, you might be able to download your content without retrieving the site's functionality. This isn't necessarily a bad choice, but you need to understand it before building your entire organisation around it.
The location of servers and backups also deserves a precise answer, especially when the site processes customer information. "In Europe" alone does not describe access, the providers involved, or the applicable commitments. Ask for the documents and the scope of the service rather than assuming compliance from a map with a flag.
Compare the service you will actually use
To compare two offers, use the same scenario: your site is online, an update causes a problem, and the previous version needs to be restored. Who intervenes? Is it included? What data is recovered? What is left to check?Then add normal operations: renewing the domain, applying patches, increasing resources, transferring the site. You will quickly see if the displayed price matches your needs or simply the rental of a piece of infrastructure.
At Cadarsir, I prefer to define these responsibilities right from the start of the project rather than letting you discover them during an incident. Truly included monitoring must remain explicit; not all sites require the same setup.
And if your problem is primarily a slow site, start by looking at what the browser experiences. A better server can help. It won't make a page light if it sends too many images and scripts. Choosing hosting is about finding a suitable base and people capable of operating it, not looking for a machine that will fix everything on its own.
