Details anonymized at the customer’s request. Pattern is real; names and exact scale are not.
Context
A regional service provider operated a long-serving core that was out of capacity headroom and expensive to spare. Subscriber traffic was continuous. The SLA effectively required the upgrade to feel like a non-event externally.
- Legacy core with limited scale and aging FRUs
- Mixed edge and aggregation layers that could not all move in one night
- Small NOC team — no luxury of a 20-person war room
Constraints
- Near-zero customer-facing downtime
- Rollback must be real, not theoretical
- Configs owned by the carrier — vendor/partner executes, customer approves
- Hardware mix of new and carefully qualified secondary units for non-unique roles
Approach
1. Lab staging first
Identical (or role-equivalent) hardware was built in a staging environment. Routing policy, IGP/EGP adjacencies, ACLs, QoS, and management access were applied and peer-reviewed before the maintenance window. The production cutover was a rehearsal, not a first attempt.
2. Layered migration
- Management/OOB and tooling paths
- Edge functions that could shift with limited blast radius
- Aggregation
- Core last, with old core retained as fallback until soak criteria passed
3. Explicit go/no-go gates
Each phase had measurable checks (adjacency counts, error rates, synthetic probes, capacity headroom). Failure meant backout — not "one more tweak."
4. People and time
- Single decision-maker on the bridge
- Pre-assigned backout commander
- Engineering coverage through soak, not only through "uplinks green"
Results
- Maintenance window completed without customer-facing downtime attributable to the cutover
- Post-migration latency and headroom goals met on agreed probes
- Legacy core decommissioned only after soak, not the morning after
What actually made it work
- Staging discipline — production was not the lab
- Rollback inventory — configs, cables, and "who plugs what back" written down
- Hardware certainty — units burned in; no unboxing mid-window
- Scope control — no feature science projects during the cutover
If your core upgrade plan does not include a dress rehearsal and a named backout owner, it is still a draft.
Nettech supports lab staging, config prep, and hardware supply for carrier and enterprise migrations. Discuss a migration · Services overview