The four options at a glance
When you plan new infrastructure or the replacement of ageing servers, you usually choose between four models. They differ in who owns the equipment, who looks after the space and who looks after operation.
| Model | Hardware | Space, power, cooling | Platform management | Payment |
|---|---|---|---|---|
| Buy, own server room | Yours | You | You | Mostly capex |
| Buy and colocate | Yours | Data centre | You or a partner | Capex and monthly space rental |
| Rent dedicated servers (hosting) | Provider's | Provider | You or the provider | Monthly rental |
| Virtual data centre (IaaS) | Provider's, shared platform | Provider | Provider (platform), you (virtual servers) | Opex by resources |
No option is always the cheapest. The decision depends on the size and stability of the workload, availability requirements, skills in your team and the period you plan for. Make a rough calculation for your own environment with the TCO calculator.
When your own server room still makes sense
Buying a server and running it on your own premises is not obsolete. It makes sense mainly when:
- systems directly control production or on-site devices and need very low latency or must work without a connection to the outside world,
- you already have a room with redundant power and cooling and a team that maintains it,
- the workload is large, stable and predictable, and you use the equipment for five years or more,
- contracts or regulation require data to be physically at your location.
In all other cases it is worth checking whether your own room really still gives an advantage or just hides costs in other budget lines.
The hidden costs of your own server room
Comparisons most often set the purchase price of a server against a monthly rental. The total cost of ownership (TCO) of your own room also includes:
- electricity with cooling: a server with an average draw of 450 W uses about 3,900 kWh a year, and with cooling and losses in a smaller room (PUE around 1.8) more than 7,000 kWh,
- UPS and air conditioning: purchase, servicing, battery replacement and a spare unit for when the first fails,
- space and security: floor area, fire suppression, access control, insurance,
- labour: replacing disks, firmware updates, monitoring, out-of-hours visits during outages,
- hardware refresh: after five to seven years the equipment has to be replaced, usually in full,
- outage risk: a single power feed or one fibre link means that one fault stops everything.
For most organisations the cost of downtime is the biggest unknown. If an hour of downtime of key systems costs several thousand euros, the difference in availability between a room in the basement and a data centre quickly shows in the result.
Colocation: your servers in a data centre
With colocation you buy the servers yourself, for example from Dell or HPE, and install them in a rack in a professional data centre. The provider supplies redundant power with generators, cooling, physical security, fire protection and several independent connectivity providers. You remain the owner of the equipment and keep full control of the architecture.
Colocation is a good choice when you want to keep your own equipment and know-how but no longer want to maintain a room. Costs are predictable: a monthly price per rack unit (U) or rack, and electricity by consumption. Because modern data centres have a better PUE than smaller server rooms, the electricity bill is usually lower too. Watch the access time when something fails, so it is worth arranging a remote hands contract or managed services.
Renting dedicated servers
With dedicated hosting the physical server is entirely yours to use but owned by the provider. The provider looks after the hardware, replacement of failed parts and the space, while you usually manage the operating system or hypervisor. There is no up-front investment, and the monthly rental includes equipment, space and electricity.
The model suits workloads that need dedicated hardware for licensing, performance or isolation reasons, but where you do not want to tie up capital. In a comparison, check whether storage is included in the price, how quickly a failed part is replaced and whether you can upgrade the server at the end of the contract without another migration.
Virtual data centre (IaaS)
A virtual data centre (VDC) is a form of infrastructure as a service. Instead of servers you rent resources, that is vCPU, RAM and storage, and create virtual servers and networks in them yourself. The provider looks after the physical servers, power, cooling and high availability of the platform.
- there is no up-front investment and no hardware refresh,
- you adjust capacity on the fly, without waiting for delivery,
- the failure of an individual server is not your concern,
- you pay for reserved resources, even when you do not use them in full.
A virtual data centre is usually most competitive for small and medium environments, for variable workloads and when you have no hardware team. For very large and stable workloads, buying can be cheaper, so a TCO calculation is essential. Data location and compliance matter too: a provider with data centres in the EU makes it easier to meet requirements for keeping data within the EU.
TIER levels: what they mean for availability
The TIER classification by the Uptime Institute describes the redundancy of a data centre's power and cooling. The stated availability applies to the facility's infrastructure, not to your servers and applications.
| Tier | Essence | Availability | Longest downtime per year |
|---|---|---|---|
| TIER I | Basic infrastructure without redundancy | 99.671% | about 28.8 hours |
| TIER II | Redundant components, a single power path | 99.741% | about 22.7 hours |
| TIER III | Maintenance without shutdown, multiple power paths | 99.982% | about 1.6 hours |
| TIER IV | Fault tolerance | 99.995% | about 26 minutes |
For most business systems TIER III is a good balance between price and availability. An average company server room rarely reaches TIER II on this scale.
The hybrid model
The decision does not have to be one for everything. A common and sensible combination is:
- stable, critical or licence-heavy systems on your own servers in colocation,
- development, test and seasonal environments in a virtual data centre,
- disaster recovery copies at a second location, for example with Veeam.
A hybrid setup needs unified management, a well-thought-out network and a single approach to backup. To see how to set recovery targets, read the guide Backup, RPO and RTO. If you are also considering replacing your virtualisation platform, check the comparison Proxmox instead of VMware.
How a migration works
- Inventory. Servers, virtual machines, actual vCPU, RAM and storage use, dependencies between systems and availability requirements.
- Classification. For each system, decide the target model: stay, colocation, rental or virtual data centre.
- TCO calculation. Compare the options over the same period using actual quotes.
- Network and security. Links between sites, VPN, segmentation, administrator access.
- Pilot. Migrate less critical systems and test a restore from backup.
- Phased migration. In groups, in agreed maintenance windows.
- Handover to operations. Monitoring, documentation and responsibilities.
A migration is easiest to plan around a natural milestone, for example when a warranty, lease or virtualisation contract expires. That way you avoid double costs and duplicated work. In larger environments a gradual approach pays off: first migrate the systems with the fewest dependencies, and key applications only once monitoring, backup and recovery procedures have been tested in the new environment.
Checklist before you decide
- Do you know your actual resource use, not just the installed capacity?
- What does an hour of downtime of key systems cost?
- Who maintains the hardware when it fails outside working hours?
- When do the warranties of your existing servers expire?
- Do the quotes include electricity, storage, connectivity and backup?
- What are the requirements for data location and audit trail?
- Does the provider have a TIER III certified data centre and copies at a second location?
- How will you switch provider or model when the contract ends?
Humanfrog supplies Dell and HPE servers, provides colocation and hosting in TIER III data centres in Slovenia and a virtual data centre on its own cloud platform. As a Veeam Silver and Proxmox partner we also help with backup, virtualisation and managed services. To check the calculation against your own environment, talk to us.
