Physical media and Layer 2 convergence
In one line
Convergence starts at the wire: how fast a device detects a failure sets the floor for everything STP, aggregation, or routing can do above it.
Down detection
- Physical loss of signal / light - fastest; the interface goes down immediately.
- Protocol keepalives - needed when the physical layer stays up but the path is dead (e.g., across a media converter, DWDM transponder, or L2 provider) - the link looks up but traffic is black-holed.
- UDLD - catches unidirectional fibre faults that physical detection misses (one strand fails, link stays "up").
Interface convergence characteristics
Carrier-delay / debounce timers decide how long a device waits before believing a down event - tuned down for fast failover, but too aggressive causes flapping. Where the physical layer can't be trusted to signal failure, add BFD (see Fast convergence techniques) so detection is independent of media behaviour.
Media truthfulness and the interface event chain
Detection speed is a property of the medium and the interface model, not just a timer. Point-to-point fibre is the gold standard: loss of light is immediate, and the 802.3z/802.3ae remote-fault signalling negotiated at link-up lets the far side report a physical fault in hardware - which is why auto-negotiation stays enabled on GigE/10GigE fibre (disabling it forfeits the fault channel, a classic false economy). The medium then sets the floor: fibre linecards debounce a down event in roughly 10 ms, while copper needs on the order of 300 ms to tell failure from noise - so "fibre between layers" is a convergence decision, not merely a bandwidth one. The interface model sets the next floor: a routed port collapses the chain to link down -> interface down -> routing update (single-digit milliseconds), whereas an SVI adds autostate - the VLAN interface only falls once the last L2 port in that VLAN dies - stretching the same failure to ~150-200 ms and coupling L3 convergence to L2 membership. Anything that inserts an L2 aggregation point between nodes (hub, media converter, provider bridge) breaks direct hardware notification entirely and leaves detection to software hellos. On fibre interconnects, run UDLD aggressive: unlike normal mode (which err-disables only the end that detects the one-way condition), aggressive mode retries and then err-disables both ends, so a half-dead uplink cannot linger in the topology.
Why it matters for design
A "redundant" link only helps if its failure is detected quickly and truthfully. Media that hides faults (transponders, provider L2) is exactly where keepalives/UDLD/BFD earn their place - and where optimistic STP/routing tuning gets defeated by slow or false detection (see Factors affecting convergence).
IPv6 / dual-stack note
Detection is L1/L2 and address-family agnostic; BFD runs per address family if you need v4 and v6 sessions.
Related
- Spanning Tree - types, tuning, and loop mitigation
- Fast convergence techniques
- Factors affecting convergence
Spaced repetition
When does a link appear "up" yet black-hole traffic, and what detects it?
When the physical layer stays up across a converter/transponder/provider-L2 but the path is dead; protocol keepalives, UDLD, or BFD detect it.
UDLD specifically protects against [...].
UDLD specifically protects against unidirectional link faults (one fibre strand fails while the link still shows up).
Tuning carrier-delay/debounce too aggressively trades fast failover for [...].
Tuning carrier-delay/debounce too aggressively trades fast failover for link flapping (false down events).
Routed port vs SVI on the same physical failure - why do convergence times differ by an order of magnitude?
A routed port collapses the chain to link down -> interface down -> routing update (single-digit ms); an SVI adds autostate - it only goes down when the last L2 port in the VLAN dies - stretching the event to ~150-200 ms and coupling L3 convergence to L2 membership.
Fibre linecards debounce a down event in ~[...] while copper needs ~[...] - one reason inter-layer links should be point-to-point fibre.
Fibre linecards debounce a down event in ~10 ms while copper needs ~300 ms - one reason inter-layer links should be point-to-point fibre.
Sources
- Cisco Live BRKCRS-2031 (Cisco Plus Canada 2012): Multilayer Campus Architectures and Design Principles - remote-fault signalling, fibre/copper debounce asymmetry, routed-port vs SVI autostate, UDLD aggressive
- Cisco Press, Cisco LAN Switching; Optimal Routing Design (convergence).