IMD SolutionComputing LibraryShop main store

Checklist

Proxmox for SCADA and PLC engineering virtual machines

Remote access and virtualization Published 7 min read

Short answer

Yes — Proxmox runs Siemens SCADA and PLC engineering VMs; field teams use it for TIA Portal engineering over RDP and in-cabinet dedicated hosts. Commit per workload: pass the support-position, Layer 2 bridge, license and restore tests below before production, and size the host from measured demand.

On this page
  1. What field use proves, and what it does not
  2. Checklist: support position and workload inventory
  3. Checklist: guest build and remote access
  4. Checklist: the Layer 2 path to the controllers
  5. Checklist: licenses and physical devices
  6. Checklist: snapshots, backups and recovery
  7. Sizing the host: two Siemens guests on 32 GB
  8. The decision, case by case
  9. Questions asked

Yes — Proxmox runs Siemens SCADA and PLC engineering virtual machines. Reported field use covers a Proxmox cluster hosting TIA Portal engineering alongside HMI and electrical-design software reached over RDP, industrial computers in cabinets running single dedicated guests, and a rail SCADA platform evaluated in the lab on Debian-based guests. None of that is an acceptance result for your client. Acceptance is per application: the vendor's support position, PLC communication over a Layer 2 bridge, license behavior after clone and restore, device passthrough, and recovery from backup rather than snapshot. The checklist below takes each item from untested to a written pass, and the worked example sizes the host.

What field use proves, and what it does not #

Proven in the field, as reported by engineering teams: Proxmox clusters running TIA Portal, HMI and electrical-design tools with RDP into each guest; snapshots and backups making engineering changes reversible; in-cabinet hosts running one dedicated application; and a rail SCADA system on Debian-based guests, evaluated successfully in the lab, with no hypervisor-level clustering because redundancy lives in the application. Most teams arriving at Proxmox now are leaving VMware over licensing and support access, not technology — nothing below depends on why you left.

Field use does not give you acceptance. Siemens documents the virtualization platforms it supports for TIA Portal and publishes a guide, Using TIA Portal in a Virtualized Infrastructure, that also recommends staying on current product versions. WinCC Professional runtime support is documented separately from the engineering tool. The migration split teams report is simple: a single server often stays on Hyper-V, while multi-host designs with live migration go to Proxmox — if that choice is still open, a guide to hypervisor selection and AMD CPU sizing works through the trade-offs.

For the in-cabinet pattern — one host, one dedicated guest — a fanless mini PC is the natural form factor, and the same acceptance checklist applies to it.

Checklist: support position and workload inventory #

  • Record the exact workload list. Product, version, guest OS, licensing method, physical interfaces. Pass when it matches the installed software, not the proposal document.
  • Get each vendor's virtualization position in writing. TIA Portal has a documented supported-platform list; each runtime has its own. Pass when every workload has a recorded position, including "not listed, accepted as our risk."
  • Keep versions current. Siemens' virtualization guidance recommends using the latest product versions, so plan the migration onto current releases instead of carrying old ones forward.
  • State the availability requirement per workload. With application-level redundancy, a single Proxmox host is a legitimate design; if the hypervisor must fail over, plan the multi-host cluster before the first production guest.
  • Test the machine suspected of not running Hyper-V. An untested concern is not a result. Its outcome does not decide Proxmox acceptance, but it decides the client's fallback.

Checklist: guest build and remote access #

  • Build one representative guest per workload — OS, application, drivers — and boot it clean. Pass when there are no unresolved device errors.
  • Size each guest from measured demand in the trial, then confirm under concurrent load. Pass when project load and compile times match the physical reference machine.
  • Work through the access path engineers will actually use. RDP into the guest, open a representative project, load, edit, compile. Pass when compilation completes without session or application errors.
  • Run the workloads concurrently and watch host and guest resources. A responsive engineering desktop proves nothing about a SCADA runtime competing for the same disks. If TIA Portal feels slow, treat it as resource contention first — fixing slow automation VMs on a Windows 11 host applies whatever hypervisor runs the guest.

Checklist: the Layer 2 path to the controllers #

  • Map every guest interface to its bridge, and every bridge to a physical uplink, on the network it serves — engineering, control, management. Pass when each guest address matches the approved design.
  • Go online to a known controller address with the intended interface selected in the engineering tool. Pass when the controller identity and diagnostics come back.
  • Run the discovery scan the commissioning workflow needs. Reaching a known address is routing; discovering devices is local Ethernet traffic. If online access works and discovery fails, the interface mapping or bridge uplink is wrong — not the controller.
  • Test SCADA acquisition separately from engineering access: live values, alarms and history under representative traffic.
  • Do not bridge the office and control networks, or flatten segmentation, to make discovery pass. That decision belongs to the network design, never to the hypervisor.

Checklist: licenses and physical devices #

  • Inventory every license — servers, activation methods, dongles. Pass when each has an owner and a written restore procedure.
  • Expect activation to notice virtualization. Cloning, restoring or moving a guest changes the hardware identity the license sees. Exercise activation in the trial guest, then again after a clone and after a restore.
  • Drop or relocate what cannot survive it. One field report replaces a logging product outright because annual manual reactivation made it a no-go on virtualized hosts. Make that call before migration, not during it.
  • Treat passthrough as a pin. Passing a host device — a license dongle, a programming adapter — to a guest ties that guest to hosts that can offer the same device path. Test the device in the guest and record the constraint.

Checklist: snapshots, backups and recovery #

  • Use both, for different jobs. A snapshot is a rollback point on the same storage; a backup is a restorable copy under a retention policy. Losing the storage loses the snapshot with it.
  • Handle live data correctly. A running historian database needs backup that is aware of it; reverting the whole guest can revert history, configuration and license state with it.
  • Remember the controller is not virtual. Restoring the engineering guest does not undo a program already downloaded to the PLC.
  • Prove a restore and a host move in isolation. Pass when the application comes up licensed and functional on the recovered guest.
  • Plan major upgrades as recovery events. Clone the host's boot drive before a version jump — cloning the boot drive before a Proxmox 8 to 9 upgrade is the working pattern.

Sizing the host: two Siemens guests on 32 GB #

The case: one host, a TIA Portal V18 engineering guest on Windows 11 Enterprise — a combination reported working on a virtualized workstation — and a Debian-based SCADA runtime guest, the field-reported rail SCADA pattern. The RAM allocations are illustrative planning figures; replace them with measured demand from your trial guests before purchase.

Inputs:

  • Engineering guest: 16 GB
  • Runtime guest: 8 GB
  • Host and storage reserve: 4 GB
  • Installed RAM: 32 GB

Rule in words: guest allocations plus the host reserve must not exceed installed RAM; the remainder is headroom for peak demand.

16 GB + 8 GB = 24 GB allocated to guests. 24 GB + 4 GB = 28 GB committed. 32 GB - 28 GB = 4 GB of headroom, which is 4 / 32 = 12.5% of installed RAM.

Decision: two guests fit, with headroom thin enough to watch, because compilation spikes push guests above their average use. The threshold is installed RAM, not a rule of thumb: a third 8 GB guest needs 28 GB + 8 GB = 36 GB against 32 GB installed, so it does not fit. Raise the host's RAM or give the third workload its own host before committing the design.

The decision, case by case #

  • Engineering VMs — TIA Portal, HMI and design tools over RDP: run them on Proxmox. Field use is broad, and a failed guest costs an engineer an hour, not production. Sign off when the guest-build and Layer 2 checks pass.
  • Production SCADA runtime: commit on four results. The runtime's support position recorded, acquisition proven over the bridge, a restore from backup proven, and availability in the application or in a multi-host cluster. A single host with no application redundancy means accepting an outage window — write that acceptance down.
  • The machine that may not run Hyper-V: test it now. A limit found during acceptance is a finding; the same limit found during commissioning is a delay.
  • Licensing decides the stack. Anything that cannot survive clone, restore and move gets a license server of its own or gets replaced — one team already chose replacement over administering annual reactivations.

Questions asked

Is TIA Portal supported on Proxmox?

Siemens publishes a supported-platform list for TIA Portal and a guide titled Using TIA Portal in a Virtualized Infrastructure. Check the list for your TIA Portal version. Where Proxmox is not on it, a hypervisor problem becomes yours to solve, which is tolerable for engineering guests and marginal for a production runtime. Record the position either way so the client knows exactly what was accepted.

Ping to the PLC works but the engineering tool cannot find devices. Why?

Ping proves IP routing; discovery broadcasts stay on the local Ethernet segment and only cross the full path of guest interface, bridge and physical uplink. If the guest interface sits on the wrong bridge or behind a NATed interface, routing works while discovery fails. Fix the interface mapping, and never widen the plant network to compensate.

Is a Proxmox snapshot enough before a risky engineering change?

For a reversible engineering change, yes — that is exactly how teams make TIA Portal modifications safe to roll back, and snapshots are meant to be short-lived. For data protection, no: a snapshot lives on the same storage as the guest, so losing the storage loses both. Keep a real backup on separate storage and prove a restore before you depend on it.

What happens to licenses when I clone or move the VM?

Virtualization presents a different hardware identity to the guest, so activation can fail after a clone, a restore or a host move. Test activation for all three events during acceptance, keep a written reactivation procedure per product, and put the reactivation work in the maintenance calendar so it is planned rather than discovered at two in the morning.

Keep reading

  • Troubleshooting

    Fixing a VM remote PLC connection with a NAT adapter

    Set the guest's virtual adapter to NAT with DHCP, disable every other guest adapter, and keep the remote-access soft key on the host. The failure that repeats is a second bridged adapter pulling the PLC route away from the NAT path. Test the host path first, then the guest, and check that no local subnet matches the plant network.

    Remote access and virtualization

  • Troubleshooting

    Fixing slow automation VMs on a Windows 11 host

    The usual cause is the Microsoft hypervisor layer under VMware, confirmed by the turtle icon in the status bar. Turn off Memory Integrity, disable Virtual Machine Platform and Windows Hypervisor Platform, run bcdedit /set hypervisorlaunchtype off, reboot, and rerun the same workload — with IT approval on a work machine. Guest memory is ruled out: 16 GB did not help, and working setups run smaller allocations.

    Remote access and virtualization