Hosting tiers are usually compared on specification. The more useful comparison is on who does the work, because that cost is real and it is never on the pricing page.
The four tiers, honestly described
Shared. Cheapest. You get an allocation on a server with many other customers. The provider handles everything below your application. Limits exist and are not always advertised. Right for brochure sites, small business sites and low-traffic applications.
VPS. A guaranteed slice of a machine with root access. You control the environment and you own its maintenance — updates, security, backups, monitoring. Meaningfully cheaper than dedicated, meaningfully more work than shared. A VPS nobody administers is worse than shared hosting.
Dedicated. A whole physical machine. Predictable performance with no noisy neighbours, and the strongest position for compliance requirements about isolation. Highest fixed cost, and you pay for peak capacity continuously.
Cloud. Resources provisioned and released on demand, billed on usage. Excellent for variable load and genuinely bad value for steady load — a workload running at constant utilisation is usually cheaper on a dedicated box.
The cost nobody quotes
Unmanaged hosting assumes someone competent is available to patch the OS, respond to a security advisory, notice a full disk, verify backups and handle an incident at an inconvenient hour.
If that person exists in your organisation, unmanaged is genuinely cheaper. If they do not, you have not saved money — you have deferred a cost until an incident, at which point it arrives all at once.
Managed hosting costs more per month precisely because that work is included. Compare the monthly difference against what an hour of your team's time costs, and against what a day of downtime costs.
Sizing without guessing
Most over-provisioning comes from sizing on fear rather than measurement. Look at:
- Peak concurrent users, not monthly visits
- Whether load is steady or spiky, which determines whether elasticity is worth paying for
- Database size and growth rate
- Whether the workload is CPU-bound, memory-bound or I/O-bound — disk I/O is the most commonly overlooked and the most commonly limiting
Start smaller than feels comfortable and monitor. Scaling up is a simple operation on every tier except dedicated; over-provisioning from day one is money spent every month for capacity you may never use.
Things that cost extra
- Backups, and particularly restores
- Bandwidth beyond an allowance
- Additional IP addresses
- Control panel licences
- Managed services and support tiers
- SSL beyond a basic certificate
- Migration in, and sometimes out
A sequence that avoids both mistakes
- Start on the smallest tier that plausibly fits.
- Instrument it — CPU, memory, disk I/O, response time — from day one.
- Let it run through a genuine peak.
- Move up when the data says to, not when someone is nervous.
- Re-evaluate annually. Workloads shrink as well as grow, and nobody ever downsizes without prompting.
About the author. Kislay Chandan has worked with schools since 2014. Rather than gathering requirements from the outside, he worked inside a school to see what actually goes wrong day to day, and built what he found into the software. He still works with Shemrock Glitters in Patna — most changes to the product start with something that came up there.
Gallery
