How to Design a Data Center Interconnect Link That Survives a Fiber Cut
Sep 10, 2026| How to Design a Data Center Interconnect Link That Survives a Fiber Cut
A practical guide to redundancy, route diversity and protection switching for inter-site DCI links
Most data center interconnect (DCI) plans specify two numbers well: how much capacity the link carries and how far it has to reach. Far fewer specify what the link does at 03:00 when a contractor cuts the duct. That gap matters, because on an inter-site link the recovery behaviour is designed in at build time – it is not something you can add after the fibre is already in the ground.
This guide covers only that half of the problem: redundancy, route diversity and protection switching. If you are still choosing between grey optics, coherent pluggables and a transponder, or working out whether your span fits inside its loss budget, start from our DCI solutions overview and come back here once the optics class is settled.
1. Redundancy, Diversity and Protection Are Three Different Decisions
These three words are used interchangeably in RFQs, and they should not be. Each one removes a different failure and leaves the others untouched.
Redundancy means a second set of hardware or a second path exists. Diversity means that second path does not share physical infrastructure with the first. Protection switching means something automatically detects a failure and moves traffic, within a bounded time, without an operator.
Buying a redundant chassis pair on two fibres that run through the same duct gives you redundancy with no diversity – one excavator removes both. Buying two genuinely diverse routes with no protection mechanism gives you diversity with no automatic recovery – traffic stays down until someone reroutes it manually.
Write the requirement as three lines, not one
• How many paths exist
• How independent those paths actually are
• How the switch happens, and how fast
2. What the Protection Standards Actually Define
The vocabulary here is standardised, which is helpful when comparing suppliers who describe the same mechanism in different marketing terms.
ITU-T G.808.1
05/2014, in force. Defines the generic functional models, characteristics and objectives for linear protection switching in connection-oriented transport layers – OTN, SDH and Ethernet transport. It covers trail protection and subnetwork connection protection with different monitoring options. Ring protection is specified separately, so G.808.1 describes the linear, point-to-point mechanism almost every two-site DCI link needs.
Read the Recommendation on itu.int
ITU-T G.873.1
10/2017, with Corrigendum 1 (03/2020) and Amendment 1 (02/2022). Specifies the automatic protection switching protocol and switching operation for linear protection at the ODUk level, including 1+1 and 1:n subnetwork connection protection with inherent, non-intrusive or sublayer monitoring.
Read the Recommendation on itu.int
The practical takeaway: "1+1" and "1:n" are defined architectures with a defined signalling protocol, not vendor shorthand. When two suppliers both claim 1+1, you can ask which Recommendation and which monitoring variant, and the answers become directly comparable.
3. Which Layer Should Carry the Protection
Protection can sit at the optical line, at the digital transport layer, or in the IP/Ethernet layer above. They monitor different things, and a scheme that is invisible to a given fault cannot recover from it.
| Protection Layer | Monitors | Recovers From | Will Not Detect |
|---|---|---|---|
| Optical line protection (OLP) | Received optical power on the line side | Fibre cut, connector or splice failure, gross span degradation | Faults that don't change received power, e.g. OSNR-driven bit errors |
| ODUk linear protection (G.873.1) | Digital defects and signal quality per ODU connection | Line faults plus degraded signals and transponder-side failures | Faults outside the monitored subnetwork connection |
| Router / IP layer | Link and routing adjacency state | Almost any failure below it, eventually | Nothing structurally, but detection is slower and visible to the service |
These layers stack rather than compete. The common design error is stacking them without deciding which one owns the fast switch, so a line-layer switch and a routing reconvergence chase each other during the same event. Assign the sub-second recovery to one layer, and set the layer above it to hold on long enough not to react.
4. 1+1 or 1:1
Both put a second path under the service. They differ in what that second path is doing while nothing is wrong, and that difference drives cost and optical budget.
In a 1+1 arrangement, the signal is transmitted on the working and protection fibre at the same time and the receiving end selects between them. FB-LINK's OLP cards send identical signals on both fibres and switch within 20 ms when a line fault is detected. In the 1:1 variant, the protection path is reserved and free to carry low-priority traffic under normal conditions, so recovery involves signalling to move the service across. Both variants are card-based and install into 1U, 2U or 5U DWDM frames.
Check before you choose
Optical budget – a passive power split for 1+1 comes out of the same loss budget as the span itself
Utilisation – 1+1 keeps a second fibre lit permanently; 1:1 lets it earn its cost carrying pre-emptible traffic
Revertive behaviour – decide the wait-to-restore interval, or risk a second avoidable hit during repair
5. The Diversity You Bought May Not Be Diversity
Two circuits from two providers, on paper following two routes, routinely share physical infrastructure somewhere along the path. The places this tends to happen are predictable enough to check directly.
Building entrance. Both cables enter through the same duct or entrance room – the last hundred metres is a single point of failure regardless of the route in between.
Bridges, tunnels, crossings. Geography forces convergence; distinct routes on a map often share the one crossing available.
Wholesale underlay. Two retail providers may lease capacity on the same underlying cable. Provider diversity is not route diversity.
A route restored after a cut is not always re-spliced along its original path, so diversity verified at handover can quietly disappear months later. The check that actually settles it is documentary: ask each provider for the route record and entrance point, and ask specifically whether the two paths share any structure, conduit or cable – then re-confirm after any repair event on either path.
6. What to Put in the RFQ and the Acceptance Test
Six items, each either a number a supplier can commit to or a test you can run at handover:
• Protection architecture and layer, named against the relevant Recommendation and monitoring variant
• Specified switching time, plus the trigger conditions that count as a fault
• Insertion loss added by the protection module, subtracted from the measured span budget on both paths
• Revertive or non-revertive operation and the wait-to-restore value
• Alarm and management behaviour: which events are reported, and whether a protection-path failure raises an alarm while the service is still up
• A documented diversity statement for both fibre routes, including building entrance points
The acceptance test worth insisting on is the physical one: with production-like traffic running, disconnect the working path and measure the actual traffic interruption, then restore it and measure again. A protection scheme that has never been triggered under load is an assumption, not a capability. Repeat it after any change to either route.
Protection Standards at a Glance
| Standard | Date | Scope |
|---|---|---|
| G.808.1 | 05/2014 | Generic linear protection switching model for connection-oriented transport layers |
| G.873.1 | 10/2017 | ODUk-level linear protection: 1+1 and 1:n subnetwork connection protection |
| G.873.1 Corrigendum 1 | 03/2020 | Corrigendum to G.873.1 |
| G.873.1 Amendment 1 | 02/2022 | Amendment to G.873.1 |
7. Where Our Equipment Fits
FB-LINK builds the transport side of this: OLP protection cards, mux/demux units, amplifiers and transponders that share the same 1U, 2U and 5U DWDM frames, alongside DCI and OTN transport platforms covering 100G to 3.2T per fibre pair with transponder and muxponder options.
• 100G to 3.2T DCI Product Suite
If you send us the measured span loss on both routes, the switch or router platform at each end and the recovery time your service actually requires, our engineers will map that to a protection configuration and the loss budget it leaves you – before anything is quoted.
For the optics class, span budget and platform selection that sit above this protection design, see the DCI solutions overview.