Hosting is sold with a set of numbers that all look reassuring. Some of them describe something real about the service you will receive. Others are close to meaningless. It is worth knowing which is which.

Uptime guarantees

Almost every host advertises 99.9% uptime. Translated into time, that permits roughly 43 minutes of downtime per month, or around 8.7 hours a year. 99.99% permits about 4 minutes a month. The extra nine is a substantial engineering difference.

The more important question is what the guarantee actually entitles you to. Typically:

  • Compensation is service credit, not money — free hosting time, applied to a future invoice.
  • You must claim it. Credits are almost never applied automatically; you have to notice the outage and submit a request, often within a short window.
  • Scheduled maintenance is excluded, as is downtime caused by your own site, by third-party services, or by events outside the host's control.

So the guarantee is a modest refund policy rather than a promise your site stays up. Its practical value is as a signal — a host confident enough to publish a strong figure and an automatic credit process is telling you something about their infrastructure.

Independent uptime monitoring of your own site is more useful than any guarantee. Several services do it free at a basic level, and it means you find out about an outage before your customers do.

What actually determines site speed

Speed comes from several places, and hosting is only one of them.

Server location. Physical distance adds latency that no amount of server power removes. A server near your audience is one of the most effective and least discussed performance decisions. Check where a host's data centres are and choose the closest to your visitors.

Storage type. NVMe or SSD storage rather than mechanical drives. Almost universal now, but worth confirming on very cheap plans.

Available memory and CPU. On shared hosting this is about the limits discussed in the plan terms, and about how heavily the host packs accounts onto each machine — which they do not publish, and which independent reviews sometimes reveal.

Runtime version. Newer language versions are frequently significantly faster. A host stuck on old versions is costing you performance for free.

Caching. Server-level caching makes a large difference for dynamic sites. Whether the host provides it, and whether it requires configuration, is worth asking.

A content delivery network. Serves static assets from locations near the visitor. Often the single biggest improvement for a geographically spread audience, and available cheaply or free.

Then there is your own site, which is usually the dominant factor: image sizes, number of requests, unoptimised scripts, and heavy plugins. A slow site on fast hosting is still slow, and this is where most real gains are found.

Measuring rather than guessing

Free tools measure page load and highlight what is causing delays. Two things to keep in mind when using them.

Test from a location representative of your actual visitors. A test run from another continent tells you about latency, not about your site.

Distinguish server response time — how long the server takes to begin sending the page — from total load time, which includes downloading and rendering everything. Server response time is the part hosting controls. If that is fast and the page is still slow, the problem is in your site rather than your host.

Bandwidth

For most sites this is the least important number on the page. Ordinary websites use far less transfer than plans allow, and hosts know it — which is why it is the specification most freely advertised as unlimited.

It becomes relevant if you serve video or large downloads directly. In that case, using a dedicated video platform or object storage is usually both cheaper and better than serving from your hosting account.

The specifications actually worth comparing

  • Data centre locations relative to your audience
  • CPU, memory and process limits, stated in numbers
  • Storage type and inode limits
  • Runtime versions offered, and how quickly new ones are added
  • Server-level caching and whether a CDN is included
  • Backup frequency, retention, and restore cost
  • Free SSL, and whether it renews automatically
  • Support channels, hours, and whether real technical staff answer
  • Renewal pricing

Those tell you what the service will be like. The uptime percentage on the homepage mostly tells you what their marketing department chose to print.