Each of the four protects a different object — RSTP protects the topology from loops, ERPS protects a ring, LACP protects a link, and dual homing protects a device's uplink — so the right choice depends entirely on what you are trying to keep alive.
Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches.
RSTP converges in 1-3 s, ERPS in 20-50 ms on a small ring, LACP redistributes in 200-300 ms, and dual homing takes 0.5-3 s. Those figures are not interchangeable.
LACP does not make a single TCP flow faster. A 2 x GE bundle still carries about 1 Gbps for one flow, because the hash pins every packet of that flow to one member.
A 200 ms target cannot cover an uplink failure on a LACP bundle. Split the acceptance criteria by scenario before you sign.
Combining mechanisms works only when each one owns a separate VLAN or layer. Two loop-breaking mechanisms over the same VLAN usually produce an outage, not extra resilience.
Before comparing convergence numbers, decide what is being protected.
RSTP / MSTP — there is a physical loop in the network and something has to break it in software while keeping the ability to restore it. What is protected is the network against a broadcast storm.
ERPS — the topology is a ring and a single link failure must be healed in tens of milliseconds. What is protected is service continuity on the ring.
LACP — two devices are joined by several links and those links should behave as one logical link, adding bandwidth and surviving the loss of a member.
Dual homing — one switch (or a group) reaches two core devices over two independent uplink paths. What is protected is the uplink and the device behind it.
They can be stacked, but the stacking order and the boundaries have to be designed — otherwise the mechanisms interrupt each other. Network redundancy built from all four is normal in a plant; redundancy built from the wrong one becomes a permanent source of outages.
RSTP was standardised in IEEE 802.1D-2004 and changed the underlying mechanism from timer-driven to handshake-driven: ports negotiate into forwarding through Proposal/Agreement instead of waiting out the Listening and Learning timers. Typical convergence dropped from the 30-50 s of legacy STP to 1-3 s.
MSTP, introduced with 802.1Q, adds multiple instances on top of RSTP, so different VLANs can map to different spanning-tree instances. That allows genuine load sharing — the same link forwarding for VLAN 10 while blocking for VLAN 20.
The cost sits in two places. First, the entire network must live in one spanning-tree domain, and a single misconfigured edge device — an old switch that does not speak MSTP, for instance — can create a loop rather than remove one. Second, convergence slows as the topology grows; a network of several dozen switches can approach 10 s in the worst case. For production control traffic that is usually unacceptable.
ERPS — Ethernet Ring Protection Switching — is defined by ITU-T G.8032 and exists purely for ring topologies. One node on the ring acts as RPL Owner and blocks the protection link; every node sends R-APS messages periodically over a control VLAN, and when a link fails the Owner unblocks the protection path so traffic recovers in milliseconds.
The standard target is 50 ms, and the conditions for reaching it are not generous:
The ring must hold a limited number of nodes — most implementations advise 16-20 at most, because message processing time grows with node count.
No unexpected traffic bursts on the ring; R-APS messages need priority.
A single ring is simple to configure. Tangent and intersecting rings raise complexity sharply, because sub-rings and control VLANs have to be planned.
Above a certain node count, split into several rings or introduce a hierarchical structure.
Note the boundary: ERPS protects the ring, and nothing outside it. The uplink from the ring to the core needs its own protection mechanism — which is exactly where LACP or dual homing comes in.
The other boundary is the recovery mode. In revertive mode traffic returns to the original path once the fault clears, and that switch-back causes a second, smaller disturbance. In non-revertive mode traffic stays on the protection path, which leaves the ring permanently unbalanced. Production networks normally run revertive, but then acceptance has to measure two switchovers — on failure and on recovery.
LACP is defined in IEEE 802.1AX (originally 802.3ad) and aggregates several physical links into one logical link:
Up to eight active member links, plus up to eight standby members.
Traffic is distributed by hash, typically over a combination of source and destination MAC, IP address and port number. Every packet of one TCP flow hashes to the same result, so a single flow always travels one member link.
When a member fails, traffic redistributes onto the survivors within a few hundred milliseconds, with no protocol negotiation involved.
The most common misunderstanding at the selection stage: a 2 x GE bundle does not give a single flow 2 Gbps. A large file transfer between two servers still runs at roughly 1 Gbps, because it is one flow on one member. Only concurrent flows — many clients, many sessions — show the aggregate.
A second, quieter misjudgement: after aggregation, losing one link does not drop the service, but it does drop a few dozen packets. For video surveillance that is invisible; for an industrial control protocol it can trigger a timeout and a reconnect. Write that into the acceptance criteria per application, not per port.
Dual homing means one switch reaching two core devices over two independent paths. There are two ways to build it:
Dynamic routing — OSPF or RIP recalculates after the primary path fails. Convergence depends on the tuning; OSPF can reach under a second with aggressive timers, but with defaults it usually sits in the 3-10 s range.
VRRP plus static routes — VRRP (RFC 5798) provides gateway redundancy alongside static routes and interface tracking. Sub-second switching is achievable, but only if the topology is simple enough; otherwise asymmetric return paths appear.
The value of dual homing is that it protects an entire device rather than a link. Switch redundancy at device level — a second chassis, a second power feed, a second uplink path — is something LACP cannot provide: a switch that loses power, hangs its management plane or loses an uplink module is beyond its reach, while dual homing can survive it. The price is the slowest convergence of the four, plus a dependency on the core side: if the two core devices are not interconnected, switchover leaves a brief one-way black hole.
The scenario: eight industrial switches form a single ERPS ring. One of them, the aggregation point, uplinks to the core over a 2 x GE LACP bundle, and the core side runs VRRP between two chassis. The business requirement says no more than 200 ms of interruption for any single point of failure.
Break the budget down item by item.
Step one — ring link failure. ERPS, single ring, eight nodes: measured switchover lands in the 30-50 ms band, so budget the upper bound, 50 ms.
Step two — one member of the uplink bundle fails. LACP redistributes traffic onto the remaining members, budgeted at 200-300 ms. This case does not stack with the ring case, because an uplink failure does not trigger a ring switchover. Budget 300 ms — already above the 200 ms target.
Step three — find the bottleneck. The 200 ms figure cannot cover an uplink member failure in this design. Three options:
Replace the uplink with dual homing — two independent paths to two cores, OSPF tuned under a second. Slower, so ruled out.
Keep LACP on the uplink and split the acceptance criteria by scenario: ring failure ≤200 ms, uplink member failure ≤500 ms. This is what most projects actually do.
Move to a 4 x GE bundle so a single failure leaves three members. Redistribution time is unchanged, but the number of lost packets falls.
Step four — check end to end. Ring failure: ERPS 50 ms, plus about 10 ms for MAC and ARP table refresh on the last switch, plus 20-50 ms of application retransmission, giving 80-110 ms — inside the 200 ms requirement. Uplink failure: LACP 300 ms plus 0-1 s of core routing convergence, so worst case is close to 1.3 s, which can only be written down as an accepted interruption.
The conclusion is uncomfortable but useful: a 200 ms requirement has to be defined per scenario. Promising "any single point of failure recovers within 200 ms" on a LACP uplink is not achievable, and the contract should separate the two cases before signature.
| Combination | Recommended | Why |
|---|---|---|
| ERPS ring + LACP uplink | Yes | Fast healing inside the ring, bandwidth and member protection outside it; no overlap of responsibility |
| ERPS ring + dual-homed uplink | Yes | Common where the core is a redundant pair; slower than LACP but protects more |
| RSTP and ERPS on the same network | Conditional | The ERPS-protected VLAN must be excluded from the spanning-tree domain, or the two loop-breaking mechanisms fight |
| LACP bundle forming one segment of a ring | No | Member flapping makes ERPS misread ring state |
| MSTP load sharing plus ERPS | No | One VLAN controlled by two mechanisms produces a loop or a long outage |
| Dual homing and LACP together | Yes, with design | Dual homing works at device level, LACP at link level; both need explicit convergence expectations |
| Scheme | Standard | Typical convergence | Topology | Main cost |
|---|---|---|---|---|
| RSTP / MSTP | IEEE 802.1D-2004 / 802.1Q | 1-3 s, slower at scale | Any topology with loops | One spanning-tree domain; a configuration error creates a loop |
| ERPS | ITU-T G.8032 | 20-50 ms, single ring, few nodes | Ring | Protects the ring only; multi-ring design is complex |
| LACP | IEEE 802.1AX | 200-300 ms redistribution | Point-to-point, multiple links | No per-flow speedup; both ends must match |
| Dual homing | Follows OSPF / RIP / VRRP | 0.5-3 s | Dual-core structure | Slowest convergence; needs core-side cooperation |
Define switchover time per scenario — ring failure, uplink failure and device power loss are three separate figures.
Measure at least two switchovers on site: once on failure, once on recovery in revertive mode.
Record packet loss, not just time. Losing three packets in 200 ms and losing three hundred feel completely different to the application.
Verify that the ERPS control VLAN is isolated from service VLANs, and that control traffic is neither rate-limited nor dropped.
Match LACP parameters — mode, rate, timeout — at both ends, or the bundle renegotiates forever.
In dual-homing designs, confirm the return path. "Out via A, back via B" produces a puzzling one-way outage.
Rayin's industrial switches expose ERPS, RSTP/MSTP and LACP through one consistent configuration model, and the port and uplink capabilities of the 8+4 port models are listed on the industrial switch page. On the access side, dual homing and PON protection switching for OLTs are covered in the PON access section, and the overall product line is summarised in the Rayin company profile.
One note on scope: parallel redundancy protocols such as PRP and HSR, defined in IEC 62439-3, achieve zero switchover by sending every frame twice over two paths. That is a different design with a different cost — double bandwidth and paired hardware support — and it is not a substitute for tuning the four schemes above.
With ERPS deployed, is RSTP still needed? Depends on topology. ERPS only manages the ring, so tree-shaped branches outside the ring still need RSTP or MSTP if they contain redundant links. The key to coexistence is VLAN separation: exclude the ERPS-protected service VLAN from the spanning-tree instances so one pair of ports is never controlled by both. Configuration order matters too — bring up ERPS first, then the spanning tree.
Two gigabit links are aggregated. Why does a speed test still show 1 Gbps? Because a single TCP flow hashes to one member link and stays there. Aggregation raises total bandwidth for concurrent flows and adds resilience; it does not raise per-flow rate. To verify the bundle works, run a multi-threaded test and watch whether aggregate throughput approaches 2 Gbps and whether both ends distribute traffic evenly.
ERPS is specified at 50 ms. Why am I measuring 200 ms? Three usual causes: more nodes on the ring than the implementation recommends, so control messages queue; heavy bursts on the ring delaying R-APS transmission; or a transit device that does not support G.8032 and forwards control frames in software. Count nodes first, then check peak ring traffic, then confirm every node handles R-APS in hardware.
Can dual homing converge in under 100 ms? Rarely with ordinary routing protocols. OSPF needs neighbour failure detection and an SPF run; even with Hello and Dead intervals tuned to a 3:1 ratio, real convergence stays in the seconds. Going below 100 ms means moving to a parallel redundancy scheme such as PRP or HSR under IEC 62439-3, which sends duplicate frames and achieves zero switchover at the cost of doubled bandwidth and paired devices.
How many switches fit in one ring? G.8032 does not fix a hard limit; implementation limits come from control-message processing and per-hop forwarding delay. Field experience puts a single ring at 16-20 nodes before switchover time becomes unstable. Beyond that, split into multiple rings or go hierarchical, and re-evaluate storm suppression at the same time, because more nodes amplify R-APS processing load during a fault.
About the author: Sara — Sara is a Customer Manager at Rayin with over 10 years of experience in the communications field. In her free time, she enjoys badminton and swimming.
About Rayin: Shenzhen Rayin Technology Co., Ltd. — Company Profile
