Explainer
How multiple HMI projects run on one server host
Short answer
One server host can run several HMI projects at the same time: this platform sets no software limit on running projects. What decides success is the license — two concurrent client connections included by default, extendable with a documented three-client add-on — and the host's CPU and memory headroom. Size from measured simultaneous browser sessions and controller load, never from project count.
On this page
- How many HMI projects can run at the same time on one server?
- How many people can watch at once on the included license?
- How many licensed client connections do four workstations need?
- Will the controller have enough headroom for four projects plus machine control?
- Should every workstation get its own project or share one project?
- Why does an extra browser session get refused after commissioning?
- How do you verify each workstation shows its own project and live values?
- Questions asked
Yes — one server host runs multiple HMI projects at the same time. On the WebIQ server platform, software imposes no limit on how many projects run concurrently; what binds you is the client-connection license and the host's CPU and memory. A single ctrlX CPU can therefore serve four different projects to four workstations, but only when the client entitlement covers four simultaneous browser sessions and the controller still runs machine control inside its measured headroom. This article gives the license arithmetic for the four-workstation case, the measurement that decides whether the controller or a separate host should carry the server, and the checks that separate a routing fault from a data fault from a licensing ceiling.
How many HMI projects can run at the same time on one server? #
On this platform the server imposes no software limit on the number of projects running at the same time — four projects run as legitimately as one, and the two real ceilings are the licensed concurrent client connections and the host's CPU and memory.
Stored and running are different states. Storing a project makes it available for deployment; running it starts the project's runtime services and lets clients connect. Both lists live on the same server, and nothing in the software stops several projects from running side by side.
If your mental model comes from panel HMIs, where a terminal executes the one runtime built for it — the way one FactoryTalk View ME project builds one PanelView runtime — that model does not transfer to a server host. What does transfer is the resource question: every running project adds CPU, memory and network load whether or not anyone is watching it.
How many people can watch at once on the included license? #
The default server license includes two concurrent client connections, and each connected browser session — an operator station, a service laptop, a maintenance tablet — consumes one of them no matter how many projects are running or how many user accounts exist.
Those connections are WebSocket sessions: persistent, two-way links each browser holds open for live values. The protocol itself sets no cap on connection counts — on a generic server the walls are file descriptors and memory, with reports of a couple of thousand concurrent sessions on a small virtual machine — so four sessions will never tax the technology. The license line binds first, the hardware second.
Size from simultaneous sessions, not headcount. Every screen that must stay live during production counts: each station, plus the commissioning laptop that overlaps the shift.
How many licensed client connections do four workstations need? #
Four workstations that each keep one browser open need four concurrent client connections; with two included by default you are two short, and the documented add-on that enables three additional web clients brings the total to five — one connection of spare capacity.
The arithmetic uses documented figures, not estimates:
- Required sessions: 4 workstations × 1 browser each = 4 connections
- Included by default: 2 connections
- Shortfall: 4 - 2 = 2 connections
- Documented add-on: +3 web clients, so 2 + 3 = 5 licensed connections, 1 spare
The vendor documents this add-on — license SWL-W-X**-WIQ**RUN*CL03**-NNNN — as enabling three additional web clients, and one add-on serves the whole host: the count is per server, not per project.
Count the demand honestly. Four stations plus an engineer's laptop that stays connected through the shift is five simultaneous sessions; 2 + 3 = 5 covers that with zero spare, and the next session is refused. Read the installed entitlement from the server's license status before you blame the network for a rejected browser.
Will the controller have enough headroom for four projects plus machine control? #
Project count cannot answer this — you need measured numbers, because the visualization server shares the controller's CPU, memory and network with the live control program, and the decision rule is that the control cycle must hold steady with all four clients connected at full production load.
Measure instead of guessing:
- Record baseline CPU and memory from the controller diagnostics while the machine program runs at full production load.
- Deploy and start one project, connect one client, exercise its screens, and record the delta.
- Project the total, then verify it by bringing the remaining projects and clients online one at a time.
The projection, with illustrative numbers you replace with your own: the machine program runs at 40% CPU with 800 MB of memory in use, and one project plus one connected client adds 12% CPU and 200 MB. Four projects with four clients project to 40 + (4 × 12) = 88% CPU and 800 + (4 × 200) = 1600 MB (1.6 GB). Those inputs are invented for the arithmetic, not measurements — screen complexity and update rate move the real deltas.
The decision rule is concrete: the server stays on the controller only when every client updates at production rate and the control cycle time is unchanged from baseline; if the cycle stretches, values lag or diagnostics flag sustained load, move the server. The fallback host is one of the industrial panel PCs mounted at the line, carrying the visualization while the controller keeps machine control alone. The same host-versus-controller trade-off is worked through in choosing a PC-based HMI runtime for a Modbus TCP PLC.
Controller-hosted web servers are sized against concurrent sessions and controller resources elsewhere too — the same accounting appears in PLC web server concurrent session limits.
One line stays fixed: safety functions never depend on the visualization server staying responsive, and combined load must never push the control task outside its specified cycle.
Should every workstation get its own project or share one project? #
Keep projects separate when the workstations need independently maintained screens or access rights, and share one project with workstation-specific views when the screens and tags are largely common — both run concurrently on the same host, so this is a maintenance decision, not a capacity one.
Separate projects win when different people maintain different stations, when the screen sets barely overlap, or when access must differ per station: each project deploys and updates without touching the others. One shared project wins when the tags and screens are mostly common — a single build and tag database, with navigation that opens each station's start view — at the price of deliberate access design.
Both architectures consume the same client connections, so the license arithmetic above holds either way. Do not confuse either with mirroring one interface onto several screens, which is a different problem again — running one HMI interface on multiple touchscreens from one PC covers that case.
For four distinct projects, write the map first: station, assigned project, and the address that opens it. That map is what you verify last.
Why does an extra browser session get refused after commissioning? #
A refused session with a correct address is almost always the client entitlement, not the network: once the live connections equal the licensed count, the server rejects the next browser — read the session count in the diagnostics before you touch the controller.
Distinguish the ceiling from resource pressure. An entitlement ceiling refuses a session cleanly; resource exhaustion shows up first as slow updates and stuttering screens and rarely as a refusal. The server diagnostics settle it — read the live WebSocket session count next to the host's CPU and memory figures. If the session count equals the licensed entitlement, the arithmetic above has simply caught up with you; if the count sits below it and sessions still fail, the fault is in the data path, and the verification table below applies.
The cap-by-entitlement pattern is not unique to web HMIs: multiple concurrent RDP sessions for industrial HMI access is the parallel case on panel-PC hardware, and why a KTP900 Basic Sm@rtServer refuses remote client connections is the same refusal symptom with a different entitlement behind it.
How do you verify each workstation shows its own project and live values? #
Commissioning is done only when every workstation opens its assigned project from its own address, every project's values track the controller, and the live connection count in the server diagnostics equals the stations you intended — check the three separately.
| What you see | Where to look | What it means |
|---|---|---|
| Wrong workstation opens another project | Browser address and the station-to-project map | Routing fault — fix the address before touching tags |
| One widget wrong while neighbors update | That widget's binding and its project tag | Local binding, data type or tag mapping fault |
| Many values dead in one project | That project's communication configuration | Connection state, access permission or controller mapping for that project |
| Values dead across all projects | Shared controller connection and host diagnostics | Common data-path or host fault |
| Next client session refused | Licensed count versus live session count | Entitlement ceiling — extend the client capacity |
For each project, pick one controller value you can watch without touching machine behavior — a counter, a mode word, an axis position — and confirm it tracks at every station through a full production cycle. When all four screens hold their assigned projects, all four data paths track, and the live connection count reads four, the requirement is met.
The decisions in one place: run all four projects concurrently on the host, because no software limit applies; license five concurrent client connections for four stations, from the two included plus the three-client add-on; keep the server on the ctrlX CPU only while the measured combined load leaves the control cycle unchanged, and move it to a panel-PC host the moment it does; separate projects per workstation when maintenance or access differs, one shared project when the screens are common.
Questions asked
Do two browser tabs on one workstation consume two licensed connections?
Each browser tab that keeps its own live session to the server consumes one licensed connection, so two tabs on one workstation count twice. Two tabs watching one project can get a fourth workstation refused while only three people stand in front of screens. Close duplicate tabs and read the live session count before concluding that you need more licensed capacity.
Does the license limit the number of projects or the number of tags?
The documented licensed quantities are concurrent web clients and, on some server licenses, the number of tags the runtime serves. Project count is not a licensed quantity: four small projects need no extra entitlement if the client and tag allowances cover them, while one screen-heavy project can exhaust the tag allowance on its own. Read both numbers from the installed license.
Can the client add-on license move if the server later leaves the controller?
The vendor documents the three-client add-on as usable on any target device, ctrlX CORE, ctrlX OS or ctrlX IPC, so it follows the server within that device family. Read the base server license the same way before relocating it, because that document governs where the primary entitlement is allowed to run.
What changes if one workstation needs two screens side by side?
Plan entitlements by simultaneous sessions, not by workstations. A second screen that opens its own live visualization session consumes another licensed connection, so four stations where one runs two screens need five connections, not four. Confirm that your client really runs one shared session for both screens before relying on the lower count.