How does a VPS work? Hypervisor, isolation and resources

A VPS (virtual private server) is a virtual machine carved out of a large physical server in a data centre. So how does a VPS work? A piece of software called a hypervisor, most often KVM, slices up the processor, memory, disk and network. Each VPS gets its own share, its own operating system, its own IP address and root access, all isolated from its neighbours.
It all rests on four stacked layers. This article takes them one at a time and says what each one means when you come to choose a plan.
In brief
- A VPS is a virtual machine: your system, your kernel, your rules, on a shared physical server.
- KVM turns the host's Linux kernel into a hypervisor. Containers (OpenVZ, LXC) share a single kernel instead, which means weaker isolation.
- A "vCPU" is not always a physical core, and overselling can slow your server down. "Steal time" lets you detect it.
- A dedicated IPv4 address, IPv6 and upstream DDoS protection matter as much as RAM.
How does a VPS work? The four-layer diagram
Here is the picture to keep in mind. Read it from the bottom up.
Your applications (website, game server, n8n, Nextcloud...)
┌────────────────┐ ┌────────────────┐ ┌────────────────┐
│ VPS 1 │ │ VPS 2 │ │ VPS 3 │
│ Ubuntu │ │ Debian │ │ Windows Server │
│ IP + root │ │ IP + root │ │ IP + RDP │
└───────┬────────┘ └───────┬────────┘ └───────┬────────┘
┌───────┴──────────────────┴──────────────────┴────────┐
│ KVM hypervisor (Linux kernel) │
└───────────────────────────┬──────────────────────────┘
┌───────────────────────────┴──────────────────────────┐
│ Physical server: CPU, RAM, SSD, network card │
└───────────────────────────┬──────────────────────────┘
▼
Data centre network → Internet
Reading the diagram: at the bottom sits the physical server (processors, memory, SSDs, network card). Above it, the KVM hypervisor shares those resources between several VPS, each with its own system (Ubuntu, Debian, Windows Server). Your applications run inside your VPS. On the network side, traffic leaves the physical card, crosses the data centre network and reaches the internet.
Think of a block of flats: the structure and the utility connections are shared, but each flat has its own front door, its own meters and its own letterbox. If a term trips you up along the way, the hosting glossary defines vCPU, KVM and BGP.
The KVM hypervisor: the conductor of the server
A hypervisor is the software that creates virtual machines, allocates hardware resources to them and lets them coexist without treading on each other.
Type 1 or type 2 hypervisor: what is the difference?
A type 2 hypervisor installs on top of an ordinary operating system, like the virtualisation software you might run on a desktop PC. It is handy for trying out a Linux distribution and unsuitable for hosting customers. A type 1 hypervisor runs directly on the hardware ("bare metal"), and that is what hosting providers use.
KVM needs a small caveat. It is not installed "underneath" Linux: Linux itself becomes the hypervisor. It is therefore usually classed as type 1, even if some people call it a hybrid. In practice, it behaves like a type 1.
KVM in practice: Linux kernel, QEMU and CPU extensions
Three building blocks work together.
- A Linux kernel module. KVM (Kernel-based Virtual Machine) was merged into the mainline kernel with version 2.6.20, released in early 2007. It reuses Linux's scheduler and memory management.
- CPU extensions. Intel VT-x and AMD-V let the VM's code run directly on the processor at close to native speed.
- QEMU and virtio. To the host, each VPS is just a process. QEMU emulates the devices (disk, network card), and the paravirtualised virtio drivers stop that emulation from dragging performance down.
What this means for you With KVM you get your own kernel. You can load modules, enable WireGuard, pick Windows Server, or install Docker on a Linux KVM VPS without working around the restrictions of a shared kernel.
How are VPSs isolated from one another?
Each VPS lives in a sealed envelope: it cannot see its neighbour's files, memory or traffic. Four layers of separation add up.
- Memory: each VM has its own address space, partitioned by the hardware (EPT on Intel, NPT on AMD).
- Processor: the host schedules compute time between the VMs.
- Disk: each VPS has its own image or volume, invisible to the others.
- Network: a separate virtual interface, relayed through a bridge with filtering.
Container virtualisation (OpenVZ, LXC) works differently:
- the kernel is shared between all customers;
- isolation is therefore weaker, and a kernel flaw affects everyone;
- you lose freedom: no custom kernel, and Docker is sometimes restricted.
No isolation is absolute. In January 2018, Spectre and Meltdown, side-channel timing flaws in processors, showed that neighbours could in theory try to read data indirectly. The fixes come through updates to the host's kernel, microcode and hypervisor. A provider that keeps its hypervisors patched therefore matters as much as the technology itself.
What this means for you Check that the plan is genuinely KVM and not a container. If the product page does not say so clearly, treat that as a warning sign.
Are a VPS's resources dedicated or shared?
The physical hardware is shared: that is the whole point of a VPS. Your allocation, however, may or may not be guaranteed, depending on the provider's policy.
vCPU, RAM, storage: what you are really renting
A vCPU generally corresponds to one thread of the host processor (a hyperthread), not necessarily a full physical core. "4 vCPUs" can therefore mean two real cores with hyperthreading. There is nothing wrong with that, as long as those threads are not sold several times over.
| Resource | What can be guaranteed | Risk if oversold |
|---|---|---|
| vCPU | A thread reserved for your VM | Compute time stolen by neighbours |
| RAM | Memory reserved for your VM | Host-side swapping, sluggishness, even VM shutdown |
| SSD | Capacity, sometimes throughput and IOPS | Unstable disk latency at peak times |
Overselling and the "noisy neighbour" effect
Overselling (overcommit) means selling more vCPUs or RAM than physically exist, on the bet that not all customers will push their load at the same moment. The bet holds until someone maxes out. When a neighbour does, your game server lags or your website answers slowly.
You can check this yourself. Run top and look at the %st value (steal time) on the CPU line: it is the share of time your VM wanted to compute but the hypervisor served someone else. vmstat 1 5 shows it too, in the st column.
top
# %Cpu(s) line: ... 0.3 st
vmstat 1 5
# "st" column on the far right
A steal time close to zero is normal. If it stays high over time, especially when your own load is light, the host is probably overloaded. Test at several times of day: a one-off spike proves nothing.
What this means for you Two plans at the same price can hide opposite policies. OmegaCloud uses strict KVM virtualisation with no overselling, so the resources listed are genuinely guaranteed. That is what the high-performance Linux VPS offers.
How does a VPS get its IP address and network access?
The chain is short. The VPS's virtual network card (virtio-net) is attached to a bridge, a virtual switch on the host. The bridge goes out through the server's physical card, then through the data centre network, which exchanges traffic with carriers via BGP. To see what sits behind that part of the infrastructure, read what a data centre is and how it works.
The IP address comes from the provider's address blocks. It is configured statically, or automatically at deployment (cloud-init or DHCP). Two situations arise:
- Dedicated IPv4: a public address just for you. These have become scarce: the RIPE NCC has run out of free IPv4 addresses, according to GÉANT.
- Shared IPv4 behind NAT: several VPSs share one address, with a limited set of ports. A game server or a mail server copes badly with that.
IPv6 completes the picture, with addresses in abundance. Outbound bandwidth is shared at the level of the host's uplink. As for anti-DDoS, where it exists, it filters upstream, before traffic ever reaches your VPS.
What this means for you Insist on a dedicated IPv4 address, IPv6 and included anti-DDoS. At OmegaCloud you get a 10 Gbit/s network, multi-homed BGP and 1 Tbit/s Anti-DDoS included in every plan.
What this means for you day to day
Because your disk is a virtual file or volume, the host can copy it, freeze it and resize it. That brings very practical benefits:
- root access and your choice of OS;
- reboots that are independent of other VPSs;
- snapshots before a risky update;
- resizing and deployment in seconds.
There are limits too. You have no direct access to the hardware, your performance is capped by the host, and nested virtualisation is not always available. If you want to run virtual machines inside your VPS, look at the Proxmox VE VPS with nested virtualisation and its conditions.
Compared with shared hosting, a VPS supplies what is missing: an isolated system you can administer yourself. Compared with a dedicated server, the difference lies in the hardware, which is entirely yours alone. To decide between the two, read our comparison of a VPS and a dedicated server.
Frequently asked questions
What is the difference between a VPS and the cloud?
A VPS is a virtual machine with fixed resources, rented by the month. "Cloud" describes more of a consumption model: resources created and deleted on demand, often through an API. A VPS can fit into that model, particularly if the provider offers an API or a Terraform provider.
How do I connect to my VPS after buying it?
On Linux, you connect over SSH using the IP address and username your provider sends you. On Windows, you use Remote Desktop (RDP).
ssh root@IP_ADDRESS
Is a vCPU the same as a real processor core?
Not always. A vCPU generally corresponds to one host thread, which is half a physical core when hyperthreading is on. What matters is that those threads are not oversold. Check it with the %st value in top.
Can another customer see my files on a VPS?
Not in normal operation. Each KVM VPS has its own dedicated volume and a memory space partitioned by the hardware. The residual risk comes from Spectre-type flaws, which is why a provider that keeps its hosts updated is worth having.
Can I install Docker or Proxmox on a KVM VPS?
Docker, yes: you have your own kernel. Proxmox needs nested virtualisation, which the provider has to enable. Check the plan's page before you order.
How much does a Linux VPS cost?
At OmegaCloud, Linux VPS plans run from €6.99 to €36.99 per month excluding VAT, with 4 to 32 GB of RAM. There is no minimum term, and discounts reach up to 15% over 12 months. The plans are currently available for pre-order.
Choosing a VPS that keeps its promises
Before you pay, run the plan through these five criteria:
- KVM virtualisation, not a container;
- no overselling;
- dedicated IPv4 and IPv6;
- anti-DDoS included;
- hosting in France, for GDPR compliance and low latency.
Our high-performance Linux VPS ticks all five: strict KVM with no overselling, dedicated IPv4 and IPv6, 1 Tbit/s Anti-DDoS included, snapshots, SSH root access, SSD storage, and Ubuntu, Debian or AlmaLinux, at the Telehouse Paris data centre. It starts at €6.99 excluding VAT per month, with no minimum term.
One thing to know before you decide: the plans are currently on pre-order, with availability announced for 1 November 2026. Once your VPS is deployed, run top at several points in the day. A %st close to zero is the best proof that your resources really are yours.