SDH will retire. The traffic it carries will not.
Migrating an operational estate off SDH is a service-by-service exercise in preserving certainty, and it starts long before the first packet node ships.
Metro and rail operators still run SDH because it earned its place. Deterministic paths, 50 ms protection switching and hard partitioning are exactly the properties that CCTV, SCADA and operational radio were engineered against. Calling the estate legacy says nothing useful. The traffic is current, the failure modes are understood, and whatever replaces it has to be at least as certain.
Most migration trouble starts with an inventory that does not exist. Circuits accumulate over decades: documented, half-documented and inherited. The first deliverable of any credible migration is a service catalogue recording each circuit's endpoints, protection class, latency tolerance and owner, proven against the live network rather than against the record.
Only then does the packet design start. MPLS-TP and Segment Routing can reproduce deterministic behaviour, but the semantics differ. SDH protection is a property of the path; packet resilience is a property of the design. Every service class needs an explicit engineered equivalent, and each equivalence has to be argued, not assumed.
Synchronisation deserves its own workstream. SDH distributed timing as a by-product of the frame structure. Packet networks do not, so PTP with the G.8275 telecom profiles and SyncE must be engineered in from the start, with holdover behaviour understood before cutover night rather than during it.
Cutover on a live estate is a design artefact in its own right. Dual-run windows, rollback points and method statements are drawn, reviewed and gated like any other deliverable. On the LU Connect estate we have carried converged GbE, MPLS, SDH and ATM transport through successive migrations with the video WAN in continuous service. The method is the reason that was possible.
The uncomfortable arithmetic is that a migration design is a larger engineering task than the original build, because the original build did not have to keep the network alive while replacing it. Fund it, gate it and staff it accordingly.