Virtualization: Part 1 — Why it’s needed and how it works

Virtualization allows you to run multiple independent virtual machines (VMs) on a single physical server. Each VM has its own operating system, its own disks and network interfaces, while the processor, memory and storage devices of the physical server are shared between them. These resources are allocated and the machines isolated from each other by the hypervisor.

How a hypervisor is organized

Modern hypervisors rely on hardware virtualization support in the CPU: Intel VT-x and AMD-V on x86, virtualization extensions on ARM. Without it, full-featured VMs either won’t run or run slowly, so the first thing to check is that support exists and is enabled in the server’s BIOS/UEFI.

Hypervisors are commonly divided into two types.

• Type 1 (bare-metal). The hypervisor is installed directly on the server instead of a regular operating system and manages the hardware itself. Examples — VMware ESXi, Xen, Microsoft Hyper-V. This option is used in data centers and server rooms.
• Type 2 (hosted). The hypervisor runs as a program inside a regular OS: VirtualBox, VMware Workstation, Parallels. It is chosen for workstations, training benches and testing.

KVM doesn’t fit neatly into this scheme. It is a Linux kernel module: after it is loaded the kernel itself becomes the hypervisor while remaining the host’s regular operating system. Therefore KVM is sometimes classified as Type 1 and sometimes as Type 2. In practice, what’s more important is this: VM performance on KVM is close to bare metal, and Proxmox VE, most cloud platforms and Russian virtualization systems are built on it.

What virtualization provides

• Consolidation. Several services that don’t need a whole server run on one physical host, each in its own VM. Fewer servers means lower costs for hardware, electricity and rack space.
• Isolation. A crash or compromise of one VM does not affect the others. Different services don’t conflict due to library versions and OS settings.
• Rapid deployment. A new VM is created from a template in minutes, without purchasing and installing hardware.
• Snapshots. The VM state is captured before an update or experiment; if something goes wrong the machine can be rolled back in seconds. A snapshot is not a substitute for a backup: it is stored on the same storage and is lost with it.
• Full backup. A copy of a VM includes the system, applications and data, so recovery reduces to launching the machine from the copy.
• Live migration. A running VM is moved to another host without stopping it, for example before servicing a server. This requires a cluster of several hosts and shared or replicable storage.
• High availability (HA). If a host in a cluster fails, its VMs are automatically started on the remaining hosts. For a reliable quorum the cluster needs at least three nodes.

Virtual machines and containers

Containers are a different isolation method and are often confused with virtualization.

There are two kinds of containers:

• system (LXC) — similar to a lightweight VM: inside is a full system with init, services and users. Proxmox VE can run them alongside VMs;
• application (Docker, Podman) — the container runs a single application with its dependencies, and the system is assembled from several such containers.

The technologies do not exclude each other. A typical setup: on the physical server a hypervisor, inside it several VMs, and within the VMs applications run in Docker. VMs provide isolation and convenient backups, containers provide reproducible application deployment.

When virtualization is not needed

• A single service fully saturates the server, so there is nothing to share.
• An application needs direct access to hardware with minimal latency. Passing devices into a VM (PCI passthrough) solves some of these tasks, but requires IOMMU support (Intel VT-d, AMD-Vi) and complicates maintenance.
• The infrastructure is entirely in the cloud: there the provider handles virtualization and you are already renting VMs.

Which part of the series to read

• Proxmox VE (part 2) — a ready-made platform with a web interface, clustering and backup, free of charge. A sensible choice for most small companies.
• VMware vSphere and ESXi (part 3) — the corporate standard of past years. After Broadcom’s acquisition — subscription-only, and sales and support in Russia have been discontinued since 2022. Read this if you plan a migration.
• KVM (part 4) — the foundation of Linux virtualization; managed via libvirt, without a ready-made platform.
• QEMU (part 5) — a device emulator that works together with KVM and also emulates other CPU architectures.
• Xen, XenServer and XCP-ng (part 6) — the oldest open-source Type 1 hypervisor and platforms based on it.
• Hyper-V (part 7) — Microsoft’s hypervisor for Windows Server infrastructure.