Nordwind Freight: Migration Triage, Latency Physics and Licence Traps
Nordwind Freight operates 52 virtual machines in two owned data centres and has committed to a 30% reduction in run-rate infrastructure cost within twelve months. A discovery exercise produced these findings. Eleven servers have had no interactive login for 14 months but are still powered on and backed up. Nine servers run a 2008 ERP module; the vendor supports it until 2027 and refuses to support it on any infrastructure other than the customer's own hardware. Six servers host Oracle databases licensed per core; the licence terms cap the number of cores the databases may run on and require the customer to notify Oracle of the hosting environment. Four servers form the real-time warehouse control system; the scanners on the shop floor tolerate no more than 3 ms round trip to the control application, and the two data centres are 6 km apart. Twenty-two servers are commodity web and application servers running a supported Linux distribution with configuration managed by Ansible. The CIO wants a migration order and a defensible answer on the four warehouse servers.
1. Classify each of the five groups using the recognised migration options (retire, retain, rehost, replatform, refactor/rebuild) and propose a migration sequence that reaches the cost target fastest.
Read the worked answer
Use the options as decisions about value and constraint, not as a ladder of sophistication.
Eleven dark servers - retire. No login for 14 months, still backed up and powered. Take a final image or archive, confirm with the business owner in writing, then switch off. This is the only group that delivers 100% of its saving with near-zero migration risk, which is why it goes first.
Nine ERP servers - retain, with a scheduled decision. The vendor forbids other infrastructure and supports the module only to 2027. Migrating them forces an unsupported configuration. Retain on-premise, and open the replacement/replatform project now so the decision lands before the support date, not after.
Six Oracle servers - replatform or rehost only after licence work. The technology is migratable; the constraint is contractual and financial, not technical (see the licensing question below).
Four warehouse control servers - retain, or move to a hybrid/edge pattern. The 3 ms budget makes a public cloud region in another city unsuitable.
Twenty-two commodity servers - rehost first, replatform later. They are already configuration-managed and supported, so a lift-and-shift is low-risk and delivers the bulk of the run-rate saving. Optimising them (managed services, right-sizing) is a second wave once they are in the cloud and instrumented.
Sequence: retire the eleven in the first month; rehost the twenty-two in months two to four; resolve Oracle licensing in parallel; retain the ERP and warehouse systems with documented exceptions and a dated review. The 30% target is realistically met by retirement plus rehosting without touching the two constrained groups, which is also the argument that protects the programme from a rushed, unsupportable migration.
2. The CIO asks why the four warehouse servers cannot simply be placed in a cloud region. Give a quantitative answer and two alternatives.
Read the worked answer
The constraint is physics, not vendor capability.
The numbers. Light in fibre travels at roughly 200,000 km/s. A 3 ms budget for a round trip therefore corresponds to about 600 km of one-way-plus-return path in ideal conditions - and that is before switching, queueing, TLS handshakes, application processing and any WAN acceleration. In practice a round trip that must complete inside 3 ms end-to-end means the control application must sit within a few tens of kilometres of the scanners, often in the same building or on the same site.
Why the 6 km between data centres is a red herring. Inter-site distance does not matter; what matters is the distance from the scanner to wherever the control application runs. If the application moves to a cloud region 400 km away, the round trip is dominated by propagation delay alone, before any processing, and the shop floor stops.
Alternative 1 - retain on-premise with a cloud control plane. Keep the latency-critical control loop on site; move the non-real-time parts (reporting, dashboards, history, spare-parts ordering) to the cloud. This is a legitimate hybrid architecture, not a failure to migrate.
Alternative 2 - edge compute on site. Run the control application on a small managed compute node at the warehouse, with cloud-based management, patching and telemetry. This gives operational consistency with the cloud estate without moving the latency-critical hop. Note the availability implication: the edge node needs a local failure mode because the warehouse must keep working when the WAN link drops.
Present the constraint to the CIO as a documented exception with a cost, not as an engineering preference - otherwise it will be reopened every quarter.
3. What must be established about the Oracle licences before any migration of those six servers begins?
Read the worked answer
Licensing is a commercial gate that can silently erase the entire business case, so it must be resolved before the first machine is copied.
Check the metric. The databases are licensed per core, and the terms cap the number of cores they may run on. In cloud environments, the mapping from physical cores to virtual CPUs and the provider's hyperthreading and core-factor rules determine whether a given instance size is compliant or an unlicensed deployment. The number and size of instances chosen must be derived from the licence entitlement, not from performance testing.
Check licence mobility. The right to run the licences on a third party's infrastructure is a distinct right from running them on your own hardware. Some agreements permit it under conditions, some require a separate authorisation, and some prohibit it outright. Read the actual agreement and any support policy that governs it.
Honour the notification clause. The terms require the customer to notify Oracle of the hosting environment. Build that notification into the migration plan; failing to do so can convert an otherwise compliant move into a breach.
Decide the commercial route before the technical one. Options are: bring existing licences with the correct instance sizing, use a licence-included cloud service, or replatform to another engine. Each has a different cost and a different migration effort, and the answer determines whether the six servers are rehosted, replatformed or retained.
Record it. Document the entitlement calculation and the notification in the migration file. The audit question arrives years later, when the engineer who made the decision has moved on.