When Does a SCADA System Need Redundancy?
SCADA redundancy is needed when losing a supervisory function would interrupt production, remove essential operator visibility or exceed the plant's agreed recovery limit. Base the decision on the consequences of losing each function, then compare those consequences with the cost and complexity of keeping it available. Before specifying two servers, establish what happens to control, alarms, commands and records when either a server or its supporting network fails.
What does the plant lose when SCADA fails?
Start with functions, not equipment. “SCADA is down” is too broad to support a design decision.
Ask operations and maintenance to describe the loss in practical terms. Can operators still see the process? Can they change a setpoint? Can the current batch finish? Can they identify an abnormal condition and follow an agreed response?
Separate the answers into four areas:
- Production: what stops, slows down or becomes difficult to restart?
- Operator awareness: which live values, alarms and equipment states become unavailable?
- Supervisory actions: which commands, recipe changes or production instructions cannot be issued?
- Records: which events, trends or production records could be missing?
For each function, agree the longest acceptable interruption and the acceptable gap in its records. These may be different: restoring a display does not, by itself, demonstrate that its missing history has been recovered.
If a failure could affect a safety-related function or an operator response credited for safety, involve a qualified assessor. Define that requirement through the safety assessment rather than treating ordinary SCADA redundancy as proof of safe operation.
Which functions justify redundancy?
Use a consequence table to decide where continuous availability matters. The following situations are illustrative, not descriptions of a particular plant.
| Function | Illustrative consequence of loss | Decision to resolve |
|---|---|---|
| Central monitoring | Operators lose an overview but retain usable local panels | Is local coverage sufficient for the agreed outage? |
| Alarm presentation | Operators cannot see alarms needed for a timely response | What independent indication or alternative response is required? |
| Recipe transfer | A running batch continues, but the next batch cannot start | How long can production wait for restoration? |
| Historical records | Live operation continues while records are unavailable | What record gaps are acceptable, and how will recovery be verified? |
| Supervisory commands | Operators cannot issue commands from the control room | Which actions must remain available locally? |
Do not reduce every consequence to lost output. Include restart effort, product held for investigation, extra operator workload and uncertainty about process conditions.
Redundancy deserves closer consideration wherever the agreed interruption is shorter than the demonstrated restoration time. Where an outage is tolerable, a tested recovery procedure may be a proportionate choice.
Which server dependencies need checking?
Draw the path from an operator action to the controller, and from a field value back to the display. Include every service that the selected design needs along that path.
Ask the designer to identify:
- Where the active and standby services run.
- How they obtain power and access storage.
- Which databases and communication services they use.
- How clients find and connect to the available server.
- Whether authentication, name resolution or licensing services are required during recovery.
- How configuration changes reach both sides.
For each shared dependency, ask: if this becomes unavailable, which required functions survive?
A pair of servers using the same supporting service should be assessed against failure of that service. Likewise, if both server instances sit on one physical host, require a separate demonstration of what happens when that host stops.
The aim is a clear boundary around the failures the design is intended to tolerate. Document anything outside that boundary.
Does the network preserve a usable path?
Review the network separately from the server arrangement.
Trace controller communications, operator connections and the link used to coordinate the servers. Identify shared switches, gateways, power supplies and cable routes. Ask which connection remains usable after each proposed failure.
Two network connections should not be accepted as independent merely because both appear on a drawing. Require the design review and test plan to address their shared dependencies.
Specify how the system handles a partial loss of communication. One operator station might lose access while another remains connected; a server might remain reachable by clients while losing access to controllers. Treat these as separate test cases.
For each case, require an unmistakable indication of unavailable or stale information. Define which commands remain permitted and how operators confirm the actual equipment state.
What should happen during failover?
“Automatic failover” needs an acceptance checklist. Ask what the operator will experience and what the process will receive.
Agree the expected behaviour for:
- Detection: which conditions trigger a transfer, and which only raise a fault?
- Connection: how do operator stations reconnect, and what interruption is acceptable?
- Data: how are unavailable values identified, and when are they considered current again?
- Alarms: what happens to active alarms, acknowledgements and event order?
- Commands: what happens to a command submitted immediately before the failure?
- Recovery: how does the repaired server rejoin, and who authorises any transfer back?
Require explicit handling of an uncertain command result. The operator should have a defined way to establish whether the action occurred before attempting it again.
Also specify behaviour when the servers lose contact with each other while retaining other connections. Ask how the design determines which server may issue commands, and how conflicting activity is prevented.
Record these requirements in the functional design specification. Acceptance should depend on observed behaviour against that specification.
Can the plant continue under local control?
Do not assume continued local control. Verify it for each operating mode that matters.
With SCADA unavailable, demonstrate whether controllers and local panels support the agreed operation. Check the current sequence, local indications, operator actions and the route to an orderly stop.
Then examine transitions. Continuing an established run is a different requirement from starting equipment, changing a recipe, recovering after a trip or restarting after power loss.
Define a degraded operating procedure that answers:
- Who takes responsibility for local operation?
- Which actions are allowed?
- Which information must be checked at the equipment?
- When must production stop?
- How is central supervision restored and confirmed?
Assess staffing as part of this procedure. If continued operation depends on people visiting several local panels, test whether that response is practical with the people normally available.
Review alarm handling alongside local control. Our guide to why operators ignore alarms provides a useful starting point for examining the information operators need.
How should redundancy be tested?
Build the test plan around the failure cases the plant has agreed to tolerate. Use a controlled commissioning window, an agreed rollback procedure and appropriate supervision.
Include server shutdown, loss of controller communication, loss of an operator connection, loss of server coordination and restoration of the failed component. Test relevant shared dependencies identified in the design review.
Observe the whole sequence: normal operation, failure, degraded operation, restoration and return to the intended arrangement.
Record evidence for each acceptance criterion:
- What operators saw and could do.
- Whether required local functions remained available.
- What happened to pending commands.
- Which alarms and records were retained or missing.
- How long each required function was unavailable.
- Whether the recovered system matched the agreed configuration.
Repeat relevant acceptance tests after changes to the redundancy arrangement. Keep the results with the operating and recovery instructions.
What common mistakes weaken the decision?
Buying a server pair before defining the consequence. Establish the required functions and interruption limits first.
Testing only a clean server shutdown. Include the communication failures and shared dependencies within the agreed scope.
Accepting a working screen as complete recovery. Verify commands, alarms and records separately.
Leaving transfer back undefined. Specify how the repaired component rejoins and how operators know recovery is complete.
Using redundancy as the entire recovery plan. Require separate evidence that the plant can restore an agreed configuration after corruption or an unsuccessful change.
When might full SCADA redundancy be unnecessary?
A single-server arrangement may be reasonable where local operation has been demonstrated, the interruption is acceptable and restoration fits within that limit.
Document the decision and its assumptions. Revisit it when production, staffing, operating procedures or reliance on central functions changes.
The useful question is not whether every plant should have redundant SCADA. It is whether this plant can tolerate the specific functions it would lose.
Frequently asked questions
Does every SCADA server need a standby?
Assess each required service against its failure consequence and recovery limit. Avoid specifying the same availability requirement for every service without a reason.
Will production continue if SCADA fails?
Establish that through testing. Demonstrate the agreed local operating modes, including transitions and recovery, before relying on them.
Is successful failover enough to accept the system?
Include degraded operation, record handling and restoration in acceptance. Require evidence for the entire sequence.
Where should a redundancy assessment start?
Start with operations and maintenance: list what becomes unavailable, what the plant can tolerate and how recovery currently works. Use that list to define the scope of a control systems assessment.