A VPS, or virtual private server, is a virtual machine running on a shared physical server that behaves like an independent server, with its own operating system, its own allocation of CPU, memory and storage, and full administrative access for the customer. Among server fundamentals, it occupies the middle ground between shared hosting and a dedicated server.
A hypervisor on the host divides the physical hardware into several isolated guests. Each guest can be rebooted, reinstalled and configured independently, while the provider manages the physical machine, power, network uplinks and hardware replacement.
How a VPS is created on shared hardware
A VPS depends on virtualization. The provider installs a hypervisor such as KVM, Xen, VMware ESXi or Hyper-V on a physical host, and the hypervisor presents each guest with virtual CPUs, virtual memory, virtual disks and virtual network interfaces.
Some providers use container-based virtualization, such as LXC or the older OpenVZ, where guests share the host kernel. This is lighter but restricts kernel choice and some kernel features. The table compares the two approaches.
| Type | Kernel | Isolation | Typical limits |
|---|---|---|---|
| Full virtualization (KVM, Xen HVM) | Own kernel per guest | Strong, hardware-assisted | Small overhead for emulated devices |
| Container-based (LXC, OpenVZ) | Shared host kernel | Weaker, process-level | No custom kernel or kernel modules |
The provider sets resource limits per guest. Storage usually lives on local NVMe drives or on networked block storage, which allows snapshots and moving a VPS to another host.
VPS resources and their typical ranges
A VPS plan is defined by a handful of core resources. The values below reflect common market offerings and vary widely between providers and price levels.
| Resource | Common range | What to watch |
|---|---|---|
| vCPU | 1 to 32 cores | Shared or dedicated cores, CPU steal time |
| Memory | 1 GB to 128 GB | Whether memory is guaranteed or overcommitted |
| Storage | 20 GB to several TB | NVMe or network storage, IOPS limits |
| Network port | 100 Mbit/s to 10 Gbit/s | Fair use rules and traffic quotas |
| IP addresses | One IPv4, often an IPv6 /64 | Cost of additional IPv4 addresses |
The single most important difference between plans is often shared versus dedicated vCPU. On shared plans, a virtual core is a time slice of a physical thread that other guests also use, and heavy neighbors appear as CPU steal time in tools such as top.
VPS compared with other server options
Choosing a VPS means accepting shared hardware in exchange for low cost and fast provisioning. The comparison below places it among its neighbors.
| Model | Isolation | Control | Provisioning time |
|---|---|---|---|
| Shared hosting | Account-level only | Control panel, no root | Minutes |
| VPS | Virtual machine | Full root or administrator access | Seconds to minutes |
| Dedicated server | Whole physical machine | Full control, sometimes hardware choice | Hours to days |
| Bare metal cloud | Physical machine via API | Full control, no hypervisor | Minutes to hours |
The terms VPS and cloud instance describe the same basic technology. The difference lies in the surrounding services: cloud platforms add autoscaling, managed networking and per-second billing, while a classic VPS is usually a fixed monthly plan with predictable pricing.
Choosing and running a VPS
A VPS is unmanaged by default: the customer is responsible for patching the operating system, configuring the firewall, securing SSH and running backups. Managed VPS offerings shift part of this work to the provider at a higher price.
| Workload | VPS suitability | Reason |
|---|---|---|
| Small to medium websites | Very good | Root access and predictable cost |
| Development and test environments | Very good | Fast creation and snapshots |
| VPN endpoints and small services | Very good | Low resource needs |
| Databases with heavy disk activity | Conditional | Depends on guaranteed IOPS |
| Latency-sensitive or licensed workloads | Often poor | Noisy neighbors and per-core licensing |
Basic server provisioning on a fresh VPS should include key-based SSH login, disabled password authentication, a host firewall, automatic security updates and off-server backups. Provider snapshots are convenient but stored on the same platform, so they do not replace independent backups.
Sizing should start from measured demand rather than guesses. Monitoring memory use, load average and disk latency for a few weeks shows whether a plan is adequate, and most providers allow vertical resizing with a short reboot, so starting small and scaling up is usually cheaper than overprovisioning from the first day.
Limitations and common misconceptions
A VPS looks like a dedicated server from the inside, but it inherits the behavior of the shared host. Several limits follow from that.
- Noisy neighbors: other guests can affect CPU, disk and network performance on shared plans.
- Overcommitment: some providers sell more virtual resources than the host physically has.
- No hardware access: custom network cards, GPUs or hardware security modules are usually unavailable.
- Single-host failure: a VPS on local storage goes down with its host unless the provider offers live migration or failover.
- Swap and memory limits: exceeding the memory allocation triggers the OOM killer inside the guest.
The short version
A VPS is like renting an apartment in a large building instead of buying a whole house. You get your own locked space, your own keys and the freedom to furnish it as you like, while the landlord looks after the building itself.
In computing terms, one powerful machine is divided into several separate virtual servers. Each customer gets a server that feels private and fully under their control, at a fraction of the cost of a dedicated machine. The trade-off is that neighbors share the same foundation.
Where it makes a difference
A small online magazine outgrows its shared hosting plan. Pages slow down during traffic peaks, and the hosting company does not allow installing a caching layer or changing the PHP configuration.
The editor moves the site to a VPS with 4 dedicated vCPUs and 8 GB of memory and 160 GB of NVMe storage. A technician installs a web server, PHP-FPM, a database and an in-memory object cache, enables a firewall that allows only ports 22, 80 and 443, and schedules nightly backups to separate object storage.
Page load times drop noticeably, and the site handles a traffic spike from a widely shared article without errors. The magazine gains full control over its stack for a predictable monthly fee, while accepting responsibility for updates and security.