Virtualization: Part 2 — Proxmox VE

Proxmox Virtual Environment (Proxmox VE) — a virtualization platform installed on a server as a ready system and managed via a web interface. It is based on Debian, runs virtual machines on KVM and QEMU, and containers on LXC. The code is licensed under GNU AGPLv3; all features are available for free.

As of September 2026 the current version is 9.2 (released 21.05.2026): Debian 13.5 “Trixie”, Linux kernel 7.0, QEMU 11.0, LXC 7.0, ZFS 2.4, Ceph Tentacle 20.2 by default. Version 9.2 introduced a dynamic load balancer for HA clusters and WireGuard as the transport for the software-defined network. Since August 2026 an official ARM64 build is published.

What Proxmox VE can do

• VMs and containers in one interface. Full-featured VMs with any OS (Linux, Windows, BSD) and lightweight LXC containers for Linux services.
• Cluster. Multiple servers are combined into a cluster and managed from one interface from any node. Live migration of VMs between nodes and high availability (HA) are supported: when a node fails its machines are restarted on others.
• Storage. Local disks, LVM and LVM-thin, ZFS with snapshots and compression, NFS, SMB, iSCSI, and built-in Ceph — distributed storage on the disks of the cluster nodes themselves.
• Backups. Built-in scheduled jobs store copies of VMs and containers on connected storage. For incremental backups with deduplication there is a separate product — Proxmox Backup Server.
• Networking. Linux bridges, VLAN, interface bonding, software-defined networks (SDN).
• Firewall at the cluster, node and individual VM level.

Subscription: what you pay for

All features — cluster, HA, live migration, Ceph, backups — are available in Proxmox VE without a subscription. A subscription pays for support and access to the enterprise repository with more thoroughly tested updates. Prices as of September 2026, per CPU socket per year:

Without a subscription updates are taken from the pve-no-subscription repository. The developers call it suitable for testing and non-critical systems, but most small installations run on it. In Proxmox VE 9 repositories are described in deb822 format. File /etc/apt/sources.list.d/proxmox.sources:

Types: deb
URIs: http://download.proxmox.com/debian/pve
Suites: trixie
Components: pve-no-subscription
Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg

The enterprise repository (/etc/apt/sources.list.d/pve-enterprise.sources) gives an error on apt update without a valid subscription, so it is disabled on servers without a subscription.

Installation

Requirements for a working server:

• 64-bit Intel or AMD CPU with virtualization support (VT-x or AMD-V, enabled in BIOS/UEFI) or ARM64;
• at least 2 GB of RAM for the system itself plus RAM for all VMs; for ZFS and Ceph add about 1 GB per terabyte of storage;
• fast SSDs, preferably with power-loss protection, and mirroring (ZFS mirror or hardware RAID with battery-backed write cache);
• several network interfaces if a cluster and Ceph are planned.

Installation is done from the official ISO image: choose the disk (for ZFS — choose a mirror immediately), country and timezone, root password, and the management interface address. After reboot the web interface is available at https://SERVER-IP:8006.

Right after installation:

1. Switch repositories to pve-no-subscription or enter a subscription key and run apt update && apt full-upgrade.
2. Create administrator accounts with separate user accounts and enable two-factor authentication instead of working under root.
3. Close port 8006 from the internet: interface access only from the internal network or via VPN.
4. Configure backup jobs and test restoring at least one VM.

Cluster

A cluster is created on the first node with the command pvecm create CLUSTER_NAME (the name cannot be changed later); other nodes join with pvecm add CLUSTER_NODE_IP. Requirements from the documentation:

• for reliable quorum and HA you need at least three nodes; in a two-node cluster a third vote can be provided by an external QDevice;
• time on the nodes must be synchronized;
• latency between nodes should be under 5 ms, UDP ports 5405–5412 must be open between them for corosync;
• cluster traffic is better placed on a dedicated network interface.

Ceph makes sense starting from three nodes and a dedicated storage network. For two–three servers without Ceph ZFS with scheduled VM replication between nodes is often sufficient.

Proxmox Backup Server

Proxmox Backup Server (PBS) is a separate system for backups of Proxmox VE VMs and containers and Linux hosts. It makes incremental backups with deduplication, Zstandard compression and client-side encryption, so daily backups take little space. Version 4.0 (August 2025) moved to Debian 13; in 4.2 (April 2026) support for S3-compatible storage became standard. PBS should be installed on a separate server or at least on separate disks: a backup on the same host as the VMs does not protect from host failure.

Migration from version 8 to 9

The planned end-of-life for the 8.x branch was summer 2026, so Proxmox VE 8 installations should be upgraded. Procedure according to the official guide:

1. Upgrade all nodes to the latest 8.4.
2. Run the check pve8to9 --full and fix everything it warns about. The script only checks the system and does not change anything.
3. If Ceph is used, first upgrade it to Squid 19.2.
4. Make backups of VMs, switch Debian and Proxmox repositories from bookworm to trixie.
5. Run apt dist-upgrade, reboot into the new kernel.

In a cluster nodes are upgraded one by one, after migrating VMs off them.

When to choose Proxmox VE

• You need virtualization on one or several of your own servers with a convenient interface and without license payments.
• You need a replacement for VMware: Proxmox VE can import VMs from ESXi.
• Your infrastructure has both VMs and Linux services that can run in containers.

Proxmox VE is less suitable if a company requires certified Russian software — in that case they look at platforms from the registry, many of which are also based on KVM.

A practical maintenance example — expanding a VM disk on Proxmox without downtime.

Common mistakes

• Web interface exposed to the internet. Port 8006 is a common target for password guessing. Access — via VPN or an allowlist of addresses.
• Two-node cluster without QDevice. If one node fails the other loses quorum and VM management is blocked.
• Snapshots instead of backups. ZFS and LVM-thin snapshots live on the same storage as the VMs.
• Consumer SSDs under ZFS and Ceph. Without power-loss protection they perform poorly on synchronous writes and wear out faster.
• Upgrading without checks. Major version upgrades only after pve8to9 --full and backups.