Proxmox Virtual Environment

Three nodes for hyperconverged Ceph and reliable HA quorum. Smaller clusters use ZFS replication instead of Ceph.

No explicit limit; Proxmox cites production clusters of over 50 nodes. Corosync needs under 5 ms latency. Larger estates are split into clusters managed by Datacenter Manager.

Yes, a standalone node with local ZFS/LVM storage; no HA or Ceph.

Two-node clusters use a Corosync QDevice (external vote on a small third machine). Ceph is not suited to two nodes; use ZFS storage replication.

Documented Ceph stretch cluster: equal nodes and monitors per site, tie-breaker node at a third site, pools size 4/min_size 2, SSD/NVMe OSDs, no erasure-coded pools.

Yes. Nodes can run guests without OSDs, and external Ceph clusters can be added as storage.

NFS, CIFS/SMB, iSCSI, Fibre Channel/iSCSI LUNs with LVM, external Ceph and ZFS over iSCSI, alongside hyperconverged Ceph.

All-flash/NVMe Ceph supported; 25 GbE or faster recommended for NVMe. Ceph Tentacle adds FastEC for small random I/O.

No deduplication for Ceph RBD VM disks; optional Ceph/ZFS compression. Proxmox Backup Server deduplicates backups.

Ceph erasure-coded pools (k+m) supported, with lower performance than replication; FastEC in Ceph Tentacle.

Default 3 replicas with min_size 2; CRUSH failure domains (host, rack, room). Do not set min_size 1.

VM snapshots including RAM, templates with linked and full clones; space-efficient on Ceph RBD and ZFS.

Ceph OSD encryption and ZFS native encryption; client-side encrypted backups to PBS. No external KMIP key manager.

Per-virtual-disk read/write limits in MB/s and IOPS, plus CPU and memory limits per guest.

CephFS for cluster storage (ISO, templates, backups). No managed SMB/NFS file-server service.

No integrated S3 object service in Proxmox VE.

External hosts can consume the Ceph cluster as a Ceph client; no managed iSCSI/NVMe-oF target.

Built-in ZFS storage replication on calendar schedules (default every 15 minutes); Ceph RBD mirroring between clusters (documented, CLI); PBS sync jobs to remote sites.

Only within a Ceph stretch cluster; no synchronous replication between independent clusters.

No DR runbooks or automated site failover. Datacenter Manager offers cross-cluster migration, not DR orchestration.

Built-in backup scheduler plus Proxmox Backup Server 4.2: incremental, deduplicated, client-side encrypted, live-restore, file-level restore, S3 backend GA.

PBS protected snapshots, verify jobs, encrypted offsite sync and tape; no WORM lock or ransomware detection in PVE.

Veeam supports Proxmox VE natively; PVE 9 adds a third-party backup provider plugin API.

No cloud recovery site. PBS can write backups to S3-compatible object storage.

SDN zones: Simple, VLAN, QinQ, VXLAN and BGP-EVPN; 9.2 adds WireGuard and BGP fabrics and route maps. IPAM/DHCP is tech preview.

Distributed Proxmox VE Firewall at datacenter, host and VM/CT level with security groups and IP sets, integrated with SDN IPAM.

ACL-based RBAC; PAM, built-in, AD, LDAP and OpenID Connect SSO; TOTP, WebAuthn and YubiKey 2FA; scoped API tokens.

Host Secure Boot compatible; UEFI Secure Boot and vTPM for guests; AppArmor for containers. No STIG baseline or drift remediation.

No FIPS 140, Common Criteria or DISA STIG attestation published by Proxmox.

Proxmox Datacenter Manager 1.x (stable since Dec 2025): fleet view, cross-cluster live migration, RBAC, Ceph monitoring, central unattended installs.

APT-based updates and major upgrades from the GUI, node by node with live migration; HA arm/disarm for maintenance. No firmware or driver orchestration.

Full REST API (the GUI uses it), CLI, scoped API tokens, webhooks and automated bare-metal installation; Terraform/Ansible via community tooling.

Built-in performance graphs, notifications, and export to external metric servers. No capacity forecasting or right-sizing.

Resource pools and per-object permissions; no service catalogue, quotas portal or chargeback.

Live migration (including local storage), HA restart (~2 min detection and failover), affinity rules, and a dynamic load balancer for HA guests in 9.2.

No integrated Kubernetes. LXC can be created from OCI images; Kubernetes runs in VMs with third-party CSI drivers.

No public-cloud edition or cloud control plane.

SPICE and noVNC consoles; no VDI broker validated by Proxmox.

PCIe passthrough with cluster-wide resource mappings; officially supported platform for NVIDIA vGPU. No packaged AI stack.

No database service.

ISO install (GUI, terminal or serial) or unattended answer-file install; cluster and Ceph wizards in the GUI. A 3-node HCI cluster is typically built in hours.

Any server meeting the requirements (VT-x/AMD-V, enterprise SSDs, redundant NICs); no certification list or Proxmox appliances.

Single nodes or 2 nodes + QDevice with ZFS replication; small footprint; Datacenter Manager for remote fleets.

Supported: Proxmox Offline Mirror keeps air-gapped systems updated from the Enterprise repositories (subscription).

Integrated import wizard for VMware ESXi VMs (tested with ESXi 6.5-8.0) and OVF/OVA import in the GUI.

Full-featured and licensed for production use under AGPLv3, using the no-subscription repository without vendor support.

Hardware-vendor neutral; subscriptions are per socket with no hardware tie. No public-cloud offering.

Software only; hardware is supported by its own vendor.

Premium: 2-hour first response for critical issues, 24/7 from 2026-10-19; until then Austrian business days 07:00-17:00 CET.

Over 2.3 million active PVE servers, 3,000+ integration partners, a large forum community; native Veeam support.

Open source, standard KVM with raw/qcow2 disks and Ceph; no licence keys gate features.