Proxmox server as a VMware alternative: hardware, migration

For many businesses, a Proxmox server is the obvious VMware alternative. Proxmox VE is open source, runs on practically any x86 hardware and comes with clustering, high availability and backup built in. This guide compares Proxmox and VMware, sets out the hardware a Proxmox server needs and describes the migration.

Why many businesses are looking for a VMware alternative

Since Broadcom acquired VMware, the licensing model has changed fundamentally. Perpetual licences have given way to subscriptions, the portfolio has been bundled into a few larger packages, and billing is per CPU core. For some environments little changes; others pay considerably more. On top of that, newer ESXi releases have dropped support for some older CPUs, controllers and network cards. Terms keep evolving, so check the current position directly with Broadcom or your partner.

What is Proxmox VE?

Proxmox Virtual Environment (VE) is made by Proxmox Server Solutions GmbH in Vienna and is based on Debian Linux. Virtual machines run on KVM/QEMU, containers on LXC. Everything is managed through a web interface that runs on every node – there is no separate management server like vCenter. The software is licensed under the open-source AGPLv3 and every feature is free to use. An optional subscription unlocks the enterprise repository with more thoroughly tested updates and, depending on the tier, vendor support.

Proxmox vs VMware: features compared

In short, Proxmox covers what small and mid-sized environments need day to day. VMware plays to its strengths in large, highly automated estates and with software certified only for vSphere.

Proxmox hardware: sizing the server

CPU cores and overcommit

Virtual CPUs can be overcommitted because VMs rarely all run flat out at once. For typical server VMs, three to four vCPUs per physical core is fine; for databases, plan closer to 1:1. Two Xeon Gold 6230s give you 40 cores, enough for well over 100 vCPUs. With Ceph, add roughly one core per OSD. In mixed clusters, set the VMs' CPU type to a common baseline, or live migration between CPU generations will fail.

Memory: don't overcommit, budget for ZFS

Don't overcommit memory: add up all VMs plus a few GB for the host. ZFS also uses RAM as a read cache (ARC); as a rule of thumb, allow a few GB plus about 1 GB per TB of storage. Ceph needs around 4 GB per OSD. In practice, 256 to 512 GB of ECC memory per node is a sensible range.

Storage: ZFS with an HBA, Ceph from three nodes

ZFS needs to see the drives directly, so a RAID controller has to run in HBA mode or be replaced by an HBA such as the LSI 9300-8i – background in our RAID levels comparison. Use enterprise SSDs for VMs; our guide to SAS, SATA and NVMe explains which interface fits. Ceph spreads data across several servers and needs at least three nodes plus a fast network: 10 GbE minimum, 25 GbE with NVMe.

Network cards

Separate VM traffic, cluster communication (Corosync), storage and migration. Corosync mostly needs low latency, ideally on its own redundant link. For VMs and migration, 2 × 10 GbE is often enough, for example with the Intel X710-DA2. For Ceph, 25 GbE is recommended, such as the Mellanox ConnectX-4 Lx. You'll find more models under network cards.

Cluster and quorum: three nodes or two plus a QDevice

A Proxmox cluster uses a majority vote to decide which nodes may carry on. With two nodes, neither has a majority after a failure, so high availability won't kick in. The answer is a third node or a QDevice: a small external machine that only acts as an extra vote. Without shared storage, Proxmox replicates ZFS volumes between nodes at fixed intervals, for example every 15 minutes. If a host fails, the VM starts on the other node from the last replica. Ceph, by contrast, needs three real nodes – a QDevice can't stand in for a server there.

Example Proxmox server configurations

In the Ceph example, three replicas leave around 11.5 TB usable out of 34.6 TB raw; plan for no more than 70 to 80 % fill. With three nodes, the Ceph network can be cabled as a full mesh without a switch, each node connected directly to the other two. You can put together all three setups in the configurator.

Migrating from VMware to Proxmox

Since version 8.2, Proxmox VE has included an import wizard that connects to an ESXi host and pulls in VMs with their disks. Alternatively, export VMs as OVF and import them from the command line, or convert individual VMDK files. In practice:

Backups with Proxmox Backup Server

Proxmox Backup Server (PBS) is a separate, also open-source product. It backs up VMs and containers incrementally with deduplication, so daily backups take up little space. Backups can be encrypted, verified on a schedule and synced to a second PBS off site. A dedicated server with plenty of bays makes good hardware for it, such as a Dell PowerEdge R740xd with 12 × 3.5-inch bays. SSDs are ideal for the datastore; with large HDD pools, a ZFS special device on SSDs speeds up the many metadata lookups.