Infrastructure5 min read
Dedicated server, virtual machine or colocation: how to choose
Three ways to run a workload in a data centre, what each one costs you in money and attention, and how to tell which fits.
Customers rarely ask for a dedicated server or a rack. They ask for a system that stays up and responds quickly. Choosing between a virtual machine, a dedicated server and colocation is a matter of matching that system to the model that suits it.
Virtual machines
A virtual machine is a share of a physical host. It is quick to start, easy to resize and inexpensive at small sizes. It suits websites, internal tools, test environments and applications whose load rises and falls.
The trade-off is that the host is shared. For most workloads that is invisible. For a database under constant heavy load, it can show up as inconsistent performance.
Dedicated servers
A dedicated server is a whole physical machine for one customer. There is no virtualisation layer unless you add one, and nobody else's workload on the hardware. Performance is consistent, and you control the machine down to the operating system.
It suits sustained heavy load: busy production applications, large databases, and hosts on which you run your own virtual machines. The provider owns and replaces the hardware; you rent it.
Colocation
With colocation you own the hardware and rent the space, power, cooling and connectivity around it. It makes sense when you already have servers, need specific hardware, or want the long-term economics of owning equipment.
You take on more: buying hardware, replacing failed parts, planning capacity. Remote hands services cover the on-site jobs, but the equipment is yours to manage.
A simple way to decide
- Is the load light or variable? Start with a virtual machine.
- Is it heavy and steady, or does it need isolation? Use a dedicated server.
- Do you own hardware, or need equipment nobody rents? Colocate it.
- Is it more than one of these? Mix them. Many setups put databases on dedicated servers and everything else on virtual machines.
If you are sizing this for a customer, collect four numbers first: expected traffic, data size, growth over the next year and how much downtime the business can tolerate. The right model usually becomes obvious once those are on paper.
Next step