IMD SolutionComputing LibraryShop main store

Comparison

Remote SCADA access through firewalls and proxies

Remote access and virtualization Published 9 min read

Short answer

Put the SCADA application and its data connections on the plant side of the firewall and carry only an authenticated VPN tunnel and a rendered remote session across it. A proxy forwards traffic under its administrator's policy and issues no licenses; session CALs are deployed and tracked on the session host. Restrict tunnel routes to that one host, license by counted people or devices, and keep direct port forwarding to an HMI host out of the design.

On this page
  1. Where should the SCADA application sit when users connect from outside the plant?
  2. Does a proxy or a VPN issue the client access licenses a remote session needs?
  3. Which remote access option works through a plant firewall?
  4. How many session CALs does a support team need?
  5. When is SSH port forwarding better than a VPN for plant access?
  6. Why not just forward ports straight to the HMI host?
  7. What do you check when the tunnel connects but the screens show stale data?
  8. Questions asked

The workable design for remote SCADA support through a customer firewall and proxy moves almost nothing across the boundary: keep the application, its tag database and its controller connections on the plant side, and carry only an authenticated VPN tunnel and a rendered remote session. That split keeps the firewall policy to one transport and one internal destination, keeps long-lived data protocols off the public path, and puts session licensing where it belongs — on the session host you administer, not on a proxy you do not. The sections below compare the five designs that get proposed for this job, work a license-count example, and end with the decision for each case.

Where should the SCADA application sit when users connect from outside the plant? #

Put it on the plant side of the firewall: the application, its tag database and its controller connections all stay inside, and only an authenticated VPN tunnel and a rendered remote session cross the boundary.

With that placement the firewall policy shrinks to one authenticated transport and one internal destination. A remote install flips it around: the application's client-server data protocol is a long-lived TCP conversation with named plant servers, so every remote PC needs routes, name resolution and a rule per protocol, and each of those rules is a standing item to re-review. Pixels are cheap to carry; control-system data protocols were never designed to cross an internet-facing boundary.

The one machine this design adds is the session host: an always-on, headless computer inside the plant that runs the SCADA application and serves the sessions. When you specify that box, the usual fit is a fanless industrial mini PC rated for continuous duty in a cabinet or server room. If the plant already runs a hypervisor, the session host is just one more guest beside the engineering machines described in Proxmox for SCADA and PLC engineering virtual machines. Either way it sits behind the firewall and stays the only internal host the remote path can reach.

Does a proxy or a VPN issue the client access licenses a remote session needs? #

No. A proxy forwards traffic under the rules its administrator sets and a VPN moves packets across the boundary; neither one issues, hosts or administers a client access license. Session licensing lives on the terminal-services host and belongs to whoever administers that host.

Name the three functions apart and the confusion in most remote-access plans disappears. A web proxy accepts traffic and forwards it by the policies its administrator configures; a plant can open a path through it, but every path ends at that administrator's desk. A VPN carries authenticated, encrypted packets across the boundary and, terminated inside, makes the remote computer behave like an intranet host for routing purposes. A client access license is a third thing entirely: for Microsoft's Remote Desktop Services it is a separate entitlement, per user or per device, deployed and tracked on the license server that belongs to the session-host deployment.

That fixes the roles before anyone edits a firewall rule: plant networking owns the tunnel and its routes, the session-host administrator owns session permissions and the CAL deployment, and the SCADA administrator owns the application licenses, which are a separate contract line again. A tunnel that connects perfectly cannot rescue a licensing rejection, and a working license tells you nothing about whether the tunnel carries routes. If the plant answers a licensing question by offering a path through the proxy, that is transport being offered as the answer to a licensing question — send it back and ask who owns the CAL deployment.

Which remote access option works through a plant firewall? #

VPN plus a terminal session is the default for utility and plant support work: one transport to secure, one host to expose, and application data that never leaves the site. The table compares the five designs most often proposed.

Design What crosses the firewall What you must administer Decision
VPN plus terminal session Tunnel setup, then one rendered session stream VPN identity and routes; session permissions; session CALs; application licenses Default for support and operator access
VPN plus remote-control tool on one console PC Tunnel setup, then control traffic to that PC Tool licensing; how it shares the console with the local operator Good for short support sessions
SCADA client installed on each remote PC Each client's own long-lived data connections to plant servers Client versions; name resolution; per-server rules; client licenses Only when a standalone client is a hard requirement
SSH port forwarding One SSH connection carrying selected TCP ports SSH endpoint and keys; one mapping per service Point fix for a single named service
Direct port forwarding to an HMI host Raw service traffic from the internet Public exposure; translation rules; service hardening; monitoring Not as a default design

Rows 1 and 2 keep every data protocol plant-side, which is why their firewall policy holds at the tunnel plus one destination. Row 3 turns every service laptop into part of the control system's data path: standing connections to named servers, a rule per protocol, and one more client version to keep aligned with the plant server. Row 4 is pure transport — it can carry a TCP service, but it licenses nothing, resolves no names and brokers no screen. Row 5 minimizes configuration and maximizes exposure, which is exactly backwards for a process network.

How many session CALs does a support team need? #

Count people and devices, then license the smaller side: five engineers who each log in from an office desktop or a shared service laptop need five per-user session CALs or ten per-device CALs, so the per-user model wins.

Microsoft's Remote Desktop Services licensing distinguishes exactly those two CAL types: a per-user CAL follows one person across any number of devices, and a per-device CAL follows one machine across any number of users. They are deployed and tracked on the session deployment's license server, on top of the server operating system licensing, and a CAL is consumed at session login — an idle VPN tunnel consumes nothing.

Worked example, with illustrative roster numbers:

  • Users: 5 support engineers. Devices: each may connect from an office desktop or a shared service laptop, so 5 users × 2 device kinds = 10 devices.
  • Per-user model: 1 CAL per person → 5 CALs; the 10 devices are irrelevant.
  • Per-device model: 1 CAL per machine → 10 CALs; the 5 people are irrelevant.
  • One planned hire: per-user becomes 6.

Decision: people roam devices in this roster, so license the people — 6 per-user CALs. The per-device model wins when many users share few machines: 12 operators rotating through 3 shop-floor PCs is 12 per-user CALs against 3 per-device. Count both sides of your own roster; the arithmetic decides the model, not habit.

When is SSH port forwarding better than a VPN for plant access? #

Treat SSH forwarding as a point tool: it carries selected TCP ports through one authenticated connection, so it reaches one named service cleanly, but it adds no session licensing, no broker and no application placement.

A local forward maps one local TCP port to one host and port reachable from the SSH server. A mapping such as local port 5050 carried to port 5051 on an internal data server opens exactly that one connection; a second service takes a second mapping, and UDP traffic does not ride this kind of tunnel at all. That one-in-one-out behavior is both the appeal — nothing is exposed beyond the mappings you typed — and the limit.

Use it for one named service between teams that already run SSH endpoints: a historian query from an engineering workstation, a diagnostics port on one server. Do not use it as the site's remote-access architecture. It has no CAL model, no screen brokering and no place for the application to live, and the SSH server itself still needs a door through the firewall — in practice, behind the same VPN rather than listening on the open internet.

Why not just forward ports straight to the HMI host? #

Port forwarding puts a process-facing computer one rule away from the open internet: every published service is a login page for anyone who scans the address, and an HMI host was built to talk to controllers, not to absorb credential guessing from strangers.

A forwarded operator station is not only a data risk. A screen that a stranger can reach is a process control surface, so for a running plant this is a safety question, and it ends the discussion of direct forwarding as a default design. CISA's good-practice guide on managing remote access for industrial control systems treats identity, transport and the services you expose as one managed path; direct forwarding hands most of that path to whoever finds the port first.

If a service genuinely must be reachable from outside, the shape changes rather than the rule: put one hardened, authenticated entry point in front and keep the application host unexposed behind it — the pattern this reverse proxy and firewall hardening guide for self-hosted services walks through end to end. The broadband edge needs the same discipline at the router itself, which the hardening guide for broadband networks, NAT routers and host firewalls covers. Then layer the path the way the plant layers its other OT network paths, as set out in why layered OT security controls protect a network path: authenticated entry, restricted routes, logged sessions, and the HMI host still talking only inside its own segment.

What do you check when the tunnel connects but the screens show stale data? #

Work outward from the session host, not from the screen: in a terminal session the application reads its data server-side, so live graphics with dead values mean the host lost its path to the controllers or data servers, not that the firewall broke the session.

  1. Prove the screen against the plant first. Pick one value that moves and compare it with an independent local indication. A responsive screen proves only that transport works; one wrong graphic is a binding fault, not a network fault.
  2. Sort local from remote. Run the same application on a plant PC beside the session host. If the local copy updates and the remote session does not, the fault sits in transport, session authorization or CAL availability. If both are stale, the host's own path to the controllers or data servers has failed.
  3. Check the host's virtual wiring. If the session host is a virtual machine, its adapter mode decides whether it can open controller connections at all; the NAT adapter fault that blocks a VM's remote PLC connection is the classic version of this.
  4. Name the owner before changing anything. Tunnels and routes belong to plant networking, session permissions and CALs to the session-host administrator, tag bindings and drivers to the SCADA administrator.

The decision rule: judge a remote-access installation by a changing value with a current timestamp, never by whether the graphics drew.

Questions asked

Can engineers just install a SCADA client on their own laptops instead?

They can, and it works until it has to be maintained. Every laptop becomes a client deployment to version-pin against the plant server, with its own name resolution, standing data connections and a firewall rule per protocol. Client seats also carry their own licensing. For support work, one session host with a rendered session keeps that maintenance on a single machine instead of on every bag leaving the site.

Does the VPN have to be a particular type or vendor?

No. The transport technology, IPsec-based, SSL-based or the plant's existing remote-access service, is plant IT's call, because they own the gateway and its policy. What the design needs from the tunnel is independent per-user authentication, access logging and routes limited to the session host. A platform-neutral IPsec tunnel and an SSL portal both meet that when configured with those three properties.

Who should own the session-license administration after go-live?

Whoever administers the session host, and it should be one named party written into the rollout agreement, usually the integrator during commissioning and the plant's IT group afterward. Proxy and firewall administrators own transport only. Record who issues CALs, who tracks the count against the roster, and who gets the call when a login fails with a licensing message.

Is the session host better as a virtual machine or a physical box?

Both work. A virtual machine rebuilds and restores quickly and fits a plant hypervisor that already hosts engineering guests; give it fixed CPU and memory reservations so concurrent sessions do not starve each other, and attach its virtual adapter to the control segment rather than behind NAT. A physical fanless box avoids competing guests and suits a small site with no hypervisor. Choose by how the site is already run.

Keep reading

  • Checklist

    Proxmox for SCADA and PLC engineering virtual machines

    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.

    Remote access and virtualization

  • 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

    Host blue screen during serial VM connection to a PLC

    A host blue screen during a VM serial session to a PLC is a Windows 10 host crash in the USB-serial and virtualization stack, not a PLC or guest fault. Capture the stop code and version stack, check whether Workstation 14 is documented for your Windows 10 build (Broadcom lists Windows 10 1903 hosts from Workstation 15.x upward), then upgrade Workstation, refresh VMware Tools, and retest before touching adapter or PLC settings.

    Remote access and virtualization