Ring protocols repair a break, but they accept a loss of traffic of between a few tens and a few hundred milliseconds during the repair. In some applications, this gap means a lost protection relay trip command or a process going to its safe state. PRP and HSR, defined in IEC 62439-3, send every frame over two separate paths at the same time; for a single link or switch failure covered by the design, there is no switchover time because the receiver uses the other copy. A common power supply failure, a failure of the end device itself or damage affecting both paths at once is outside this protection. There is also a cost. Below, we describe the architecture of the two protocols, their integration components, frequently overlooked design constraints and how we decide between MRP/RSTP and HSR/PRP.
Bumpless redundancy: which applications cannot tolerate a gap
“Bumpless” redundancy means that frames keep arriving over the other path when there is a single link or switch failure covered by the design. This requirement typically arises in the following applications:
- IEC 61850 substations: Trip and interlock messages sent via GOOSE and the sampled values on the process bus work with a time budget of a few milliseconds. A sample stream lost during ring recovery can block the protection function.
- Motion control: In synchronous axes with short cycle times, losing several consecutive cycles leads to an axis fault. PROFINET IRT uses its own solution, MRPD, for this need; on other networks the solution depends on the protocol, and HSR/PRP is an option only if the end devices support these interfaces.
- Process safety and control systems: Synchronisation between redundant controllers and critical IO traffic may require that network-related switchovers do not affect the process.
The criterion is simple. Write down the longest communication gap the application can accept (watchdog, timeout or protection function budget) and compare it with the ring protocol’s worst-case recovery time. If the gap is shorter than the ring recovery, HSR/PRP comes onto the table; if not, a ring protocol is more economical.
PRP: two LANs that cannot see each other
PRP (Parallel Redundancy Protocol) uses two completely independent networks: LAN A and LAN B. An end device connected to both networks is called a DANP (Doubly Attached Node implementing PRP). A DANP sends one copy of every frame to LAN A and the other to LAN B, and appends a 6-byte RCT (Redundancy Control Trailer) to the end of the frame. The RCT carries the sequence number, LAN identifier and frame size. The receiving DANP discards the second copy arriving from the same source with the same sequence number and passes a single frame to the upper layer.
PRP’s strengths:
- LAN A and LAN B can be in any topology; each can run RSTP or a ring protocol internally.
- The switches in the networks do not need to know PRP; because the RCT is at the end of the frame, standard switches forward the frame normally.
- A LAN being completely out of service does not affect communication; one LAN can be shut down during maintenance.
The cost is also clear: in industrial network infrastructure, hardware and cabling double. There must be no physical connection whatsoever between LAN A and LAN B. The classic mistake we see in the field is joining the two networks “for convenience” through a management switch or an engineering PC. This breaks PRP’s independence assumption and duplicates frames. Address the switch and backup topology decision together with the distance, routing and testing requirements in the structured cabling and fiber infrastructure design.
HSR: bidirectional frame duplication on a ring
HSR (High-availability Seamless Redundancy) applies PRP’s logic to a single ring. Every device in the ring is a two-port DANH (Doubly Attached Node implementing HSR). The source device sends every frame as two copies, one clockwise and the other anticlockwise, and inserts a 6-byte HSR tag after the source MAC address. Each node in the ring forwards frames not addressed to it to the next node; the destination node accepts the first copy that arrives and discards the second. Unicast frames are removed from the ring at the destination node; multicast and broadcast frames travel around the ring and are discarded by the source when they return to it.
In HSR the ring is never blocked; loops are prevented by nodes not forwarding a frame they have already forwarded a second time. When a link breaks, one copy of the frame continues to reach its destination from the other direction. No separate LAN is needed; however, every node in the ring must support HSR. A standard switch or a device that does not support HSR cannot be placed inside the ring.
RedBox, QuadBox and integrating singly attached devices
Not all devices on site have two ports and HSR/PRP support. Devices with a single network interface are called SANs (Singly Attached Nodes) and are brought onto the network with the following components:
| Component | Function | Typical use |
|---|---|---|
| RedBox | Connects SANs to an HSR ring or a PRP LAN pair; performs duplication and duplicate discarding on their behalf | Legacy IEDs, HMIs, engineering PCs |
| QuadBox | Connects two HSR rings to each other; has four ports | Redundant connection between rings |
| HSR–PRP RedBox | Connects an HSR ring to PRP LAN A and B | HSR at bay level, PRP at station level |
| Direct SAN | Connected only to LAN A or LAN B, without redundancy | Non-critical monitoring devices (PRP only) |
A frequently overlooked point: the RedBox itself is a single point of failure. For SANs behind a RedBox, redundancy only extends as far as the RedBox; if the cable between the RedBox and the SAN or the RedBox’s power supply fails, the device loses communication. For critical devices, choose models with a built-in DANP/DANH interface. And do not run the RedBox from a single power supply.
HSR, PRP, MRP and RSTP comparison table
| Criterion | PRP | HSR | MRP | RSTP |
|---|---|---|---|---|
| Recovery time | No switchover time for a covered single failure | No switchover time for a covered single failure | Bounded by profile (e.g. 200 ms) | Depends on topology |
| Network infrastructure | Two independent LANs | Single ring | Single ring | Any |
| Switch requirement | Standard switches | All nodes HSR-capable | MRP-capable | Standard |
| Bandwidth | Full capacity on each LAN | Double traffic on the ring | Normal | Normal |
| Cost | Highest | High | Intermediate | Low |
For how MRP works and the commissioning test, see the article ring topology with PROFINET MRP.
Design constraints: bandwidth, node count, time synchronisation
Bandwidth
In PRP, each LAN carries one copy of the traffic; capacity planning is done for a single network and applied identically to both LANs. In HSR, the situation is different: every frame travels around the ring in both directions. Multicast and broadcast frames, such as GOOSE and Sampled Values, travel the entire ring in both directions; as a result, every link in the ring carries the whole multicast load twice. In designs where high-rate streams such as Sampled Values will be added, calculate link capacity with this multiplier.
Node count and latency
In an HSR ring, each node adds a delay when forwarding a frame; the delay of the copy going to the farthest point of the ring increases linearly with the number of nodes. Limit the ring size according to the application’s latency budget. A practical method: calculate the worst path using the manufacturer’s per-node forwarding delay and compare it with the application’s time budget. In PRP, latency depends on each LAN’s own topology; the latency difference between the two LANs affects the duplicate discard window at the receiver, so design both LANs with a similar depth.
Frame size
The RCT and HSR tag add 6 bytes to the frame. With maximum-size frames, this can produce frames exceeding the standard 1518 bytes. Verify in the datasheet that the switches on the network accept this size. Otherwise, large frames are silently dropped and the fault only shows up in certain applications; it is one of the hardest kinds of problems to catch in the field.
Time synchronization
The IEC 61850 process bus and applications requiring precise timestamps use IEEE 1588 PTP; in substations, its IEC 61850-9-3 profile is common. Because PTP frames are modified along the path by transparent clocks, the copies arriving from the two LANs are not identical and cannot be handled with ordinary duplicate discard logic. If PTP is to be used on an HSR/PRP network, verify that the clock source, RedBoxes and end devices support PTP over HSR/PRP in accordance with IEC 62439-3.
Decision tree for the choice
- Is the communication gap the application can accept longer than the ring recovery time? If yes, MRP (PROFINET cells) or RSTP (backbone) is enough. If no, continue.
- Is the application PROFINET IRT? If yes, PROFINET’s own bumpless mechanism, MRPD, is considered first. If no, continue.
- Do most devices have dual interfaces and HSR/PRP support? If no, the number of RedBoxes and the fact that they are single points of failure are added to the cost–benefit calculation; if the number of critical devices is small, a move to models with built-in support is planned for these devices.
- Is doubling the cabling and number of switches acceptable? If yes, PRP; the LANs are built with standard switches and topology flexibility is preserved.
- Are the devices laid out geographically along a line, with no additional switches wanted? If yes, HSR; the devices form the ring among themselves. Ring size is limited according to the latency budget.
- Do the bay and station levels have different needs? Hybrid structure: HSR within the bay, PRP at station level, with an HSR–PRP RedBox between them.
Before deciding, take an inventory of which protocols the devices on site support; this inventory often determines the theoretical preference. The work does not end after commissioning either. In PRP, one LAN silently remaining faulty leaves the system without redundancy without anyone noticing. For a setup that monitors the periodic supervision frames and duplicate counters of HSR/PRP nodes, see our industrial network monitoring and fault analysis service page, and for architecture selection and design, our industrial Ethernet and factory networks service page. For newer mechanisms such as deterministic timing and 802.1CB frame replication, the article TSN standards describes the IEEE 802.1 counterpart of ideas similar to HSR/PRP. The standard itself is published by IEC.
Frequently asked questions
Does PRP require special switches?
No. Because the RCT is appended to the end of the frame, LAN A and LAN B can be built with standard switches. But verify that the switches accept frames that are 6 bytes longer and that the two LANs are never physically joined at any point.
Can a normal switch be added to an HSR ring?
Not inside the ring; every node in an HSR ring must support HSR. Standard switches and single-interface devices are connected to the ring through a RedBox.
Can HSR and PRP be used together?
Yes. A common structure is to build HSR rings at bay level and PRP networks at station level and connect them with an HSR–PRP RedBox.
Is zero interruption always better than MRP?
No. HSR and PRP cost more in terms of hardware, cabling and bandwidth. If the communication gap the application can accept is longer than the ring recovery time, MRP or RSTP is both the right and the economical choice.