Clear, jargon-free methods to find the right host for your project.
A festival, conference, or online ticketing site cannot run on "we will see on the day." Here is how to size hosting for the spike without paying for peak capacity all year.
Putting 80 MB MP3 files on the same shared host as WordPress saturates bandwidth and backups. RSS, audio CDN and dedicated platforms — a different calculation from a brochure site.
cPanel reassures beginners; Plesk unifies Windows and Linux; without a panel you gain control and lose the safety net. The right choice follows your technical autonomy — not fashion.
One shared FTP account in LastPass means zero traceability and impossible revocation. Individual accounts, SSH keys and folder-level rights — minimum hygiene before the first freelancer.
HTTP 301, registrar change, www vs apex — a bad redirect erases SEO or traps users in an infinite loop. Map before touching DNS.
Sending 10,000 newsletters from the site shared SMTP invites spam folders and suspension. Deliverability, IP reputation, and compliance are a different job than web hosting.
Dual-stack A + AAAA is enough for most new sites. No need to rebuild your infrastructure — enable IPv6 at the host, check DNS, and test mail separately.
Gravelines looks close and cheap — until Swiss users wait 80 ms more and backups replicate to the US without a data processing agreement. A region is chosen in three dimensions, never at random.
Having a backup is not a recovery plan. RTO, RPO, tested restore, and defined roles — what separates a disk copy from an organisation that survives a major outage.
Phishing, spam, flagged content — the host opens a ticket with a twenty-four-hour deadline. Without an internal procedure, you lose the site before finding the compromised account.
Outage, data breach or illegal content — everyone points elsewhere. Without a written matrix, the client assumes "everything is included" and the host points to "your application".
"Best effort" backups, suspension without notice, data ownership — hosting terms are read on signing day, not on unpaid invoice or dispute day.
Hugo, Eleventy or plain HTML — little server, small attack surface. Until you need auth, search or a thousand pages per day. Where static holds — and where it breaks.
Static builds plus a CDN speed delivery — but search, accounts, checkout, and editorial preview stay dynamic. Map those flows before celebrating the Git deploy.
Strapi, Directus or Contentful separate content and rendering — useful with multiple fronts and autonomous front team. Otherwise you pay for two stacks for a brochure site.
A notebook running locally on 32 GB RAM is not team infrastructure. Sharing, auth, sensitive data and GPU cost — how to industrialise without Jupyter becoming shadow IT.
Hourly or monthly, spot or reserved, real or shared VRAM — a badly chosen GPU instance costs more idle than used. Ask these questions before signing.
Renting a GPU before defining latency, context, and confidentiality leads to GPU bills and a POC that never reaches production. Map usage, data, and budget first.
A vague message reassures nobody; a false ETA destroys trust. In an incident, structure updates — impact, scope, actions, next checkpoint — without guessing the cause.
A status page built in five minutes after an outage helps nobody. Prepare components, subscribers and runbook before the first 502 — your customers and support will thank you.
Four nines means 52 minutes of downtime per year — in theory. In practice, SLAs exclude maintenance, DNS, and sometimes upstream network. Read the contract, not the badge.
LCP, INP and CLS depend as much on your theme as on the server. Before switching hosts for a PageSpeed badge, separate what belongs to CDN, PHP and the front end.
HTTP/3 and QUIC promise lower latency on unstable networks. On a site already fast over HTTP/2 with healthy TTFB, gains are often marginal — sometimes negative if origin stays slow.
OPcache stops recompiling PHP on every request — immediate CPU and latency gain. Misconfigured or disabled after deploy, it stays invisible until the next outage.