A ring network keeps a whole industrial site running even when one cable is cut: switches are wired head-to-tail into a closed loop, and when any segment fails the traffic simply takes the other way around. The loop alone does nothing, though — what actually earns you uptime is the ring protocol that blocks one link by default and opens it only on a fault.
A ring topology reroutes around a broken segment instead of going dark; the protection comes from the ring protocol, not from "connecting in a circle."
ERPS (ITU-T G.8032) restores service in under 50 ms; RSTP is universal but usually needs a second or more.
A ring only covers single-link failures. You still need a redundant link margin, dual supplies, route planning, and BPDU guard.
A switch forwards a broadcast or unknown unicast frame out of every other port. Close a physical loop and that frame circles forever. With every lap more switches refresh their MAC table, pinning the same address to different ports, the table thrashes, the CPU saturates, and within seconds the whole network is down.
So connecting switches in a ring without protection gives you the opposite of redundancy — it gives you a broadcast storm that takes everything with it. The ring only works once that forwarding loop is broken on purpose.
The ring protocol designates one link as the blocked link, the RPL (Ring Protection Link), and holds it open logically. The loop becomes a simple chain, the storm disappears, and the two endpoints of the blocked link keep exchanging protocol messages. When a break occurs elsewhere, the blocked port opens instantly and traffic reroutes. That spare path is a redundant link held in reserve, not wasted bandwidth.
What decides "open or stay closed" is how the protocol message travels around the ring. ERPS sends R-APS messages in both directions at once; RSTP passes BPDU frames hop by hop; MRP uses Media Redundancy test frames. Different protocols mean different message speed and different limits on how many nodes the ring can hold.
| Protocol | Standard | Typical recovery | Nodes per ring | Best for |
|---|---|---|---|---|
| RSTP | IEEE 802.1w | Over 1 s | ≤ 7 hops advised | Office networks, interruption-tolerant monitoring |
| ERPS | ITU-T G.8032 | Under 50 ms | ≤ 16 nodes advised | Industrial sites, transit, campus backbone rings |
| MRP | IEC 62439-2 | Under 200 ms | 50 nodes max (spec) | PROFINET lines, fieldbus |
Rapid Spanning Tree Protocol (RSTP) is the most portable — nearly every commercial and industrial switch supports it, and configuration is simple. The cost is slow convergence: with many nodes, BPDU-by-hop messaging plus topology recomputation pushes recovery past one second. Video stutters for a beat or two; a PLC running PROFINET IO can drop off the bus entirely.
ERPS was built for rings. Its target switchover is under 50 ms, announced quickly by R-APS messages. Version 2 supports intersecting and sub-rings, which suits campuses, transit, and mines where a dozen-plus devices hang on one ring.
MRP lives inside the PROFINET world. Recovery is typically under 200 ms, and configuration splits roles into "ring manager" and "ring client," tied fairly tightly to the PLC side.
Break the switchover into four stages — fault detection, protocol-message propagation, opening the blocked port, and flushing the forwarding table — and the first two eat most of the budget. The last one is the easiest to forget.
Warning
If the switch does not flush its MAC table after a switchover, the 50 ms figure can quietly become 300 seconds. The table still thinks "this MAC is on the old port" and keeps sending frames into a dead link until the entry ages out — and the default aging timer is 300 seconds.
ERPS carries a flush trigger inside its R-APS message, so receivers clear learned MACs and relearn immediately. That is a big reason it restores service (not just the link) inside 50 ms. RSTP also sends a topology-change notice, but vendor implementations differ widely; if you see a site that needs a minute or two to come back after a switchover, this stage is usually the culprit. The ARP table behaves the same way — a gateway-side ARP aging near 240 seconds adds its own delay. Check the MAC table first, then the ARP table.
ERPS at 16 nodes per ring is an engineering recommendation, not a hard ceiling. Every extra node adds one R-APS hop, so recovery time climbs roughly linearly; and with only two paths between any two points on the ring, bandwidth planning and failure domains get hard to slice.
Past a dozen nodes, the common move is to split into several rings and hang the sub-rings off a main ring by dual-homing or intersection (ERPS v2 supports this). Another approach keeps the ring for access only and runs dual uplinks to two cores at the aggregation layer, with link aggregation and RSTP as the backstop. Rayin industrial switches support ERPS ring networking, and the industrial Ethernet switch solution page shows measured node counts and switchover times from real configurations.
A ring covers a single-link fault. It does not cover:
A device losing power entirely. That point becomes a break unless the device supports bypass, or carries no traffic that must stay up.
Double breaks. Two cuts at once split the ring into two disconnected chains. When planning, deliberately separate the two segments that share one cable trench or one pole.
Storms from outside the ring. Plug in an unmanaged switch with the spanning tree protocol disabled and loop it to itself, and the storm still pours in. Edge ports need BPDU guard and storm control.
Power problems. The ring does nothing for supply. For no power loss you need dual supplies or a UPS — Rayin industrial switches support 1+1 redundant power so a supply failure does not drop the service.
What is the difference between a ring and link aggregation (LACP)?
Link aggregation bundles two physical links into one logical link; it handles bandwidth and redundancy between two points. A ring handles rerouting when any segment among many points breaks. They are used together constantly: aggregate between ring nodes for bandwidth, run ERPS on the ring itself for redundancy.
Do all switches on the ring need the same protocol?
Every node on the ring must run the same ring protocol. Mixing RSTP and ERPS lets ERPS's blocked port be misjudged and opened by the spanning tree protocol, creating a logical conflict. If you must mix, keep the unsupported device outside the ring and hang it off two ring nodes with dual uplinks.
How many frames does a camera lose on switchover?
With ERPS recovering under 50 ms, a 4 Mbps, 25 fps camera drops roughly one or two frames — invisible to the eye. An RSTP switchover at the one-second level drops around 25 frames, the recording shows a clear gap, and the NVR usually reconnects once.
Is a ring really cheaper to cable than a star?
In linear scenes — highways, tunnels, mine drives — a ring saves fiber because no node needs its own run back to the center. Inside a building the ring can cost more, because that last closing link often has to loop back around.
Written by Sara, Customer Manager at Rayin — 10+ years in industrial communications. She helps integration engineers choose and deploy switching and PON equipment.
Rayin: About Rayin → /Profile/
Connect on LinkedIn
