[ Migration paths ]
What needs to move, and where?
The framework starts with your workloads. We compare public cloud, private cloud and on-premises options against cost, dependencies, data location and your team's ability to run them.
- 01Path
Move into cloud
From on-premises infrastructure to public or private cloud.
Check which applications can move as they are, which dependencies need to change and what the receiving environment must provide.
- 02Path
Move out of public cloud
Bring selected workloads back to private cloud or your own infrastructure.
Compare the cost of staying with the cost and effort of moving. Keep cloud-native services where replacing them would outweigh the benefit.
- 03Path
Move between platforms
Switch providers or consolidate virtualisation and Kubernetes estates.
Map the differences in identity, networking, storage and platform services before committing to a cutover.
- Step 01
Discover
Inventory applications, owners, data and dependencies. Record current cost, performance and recovery needs.
DeliverableWorkload and dependency map
- Step 02
Compare & plan
Compare viable destinations with staying put. Classify each workload: move as is, adapt, rework or retain.
DeliverableTarget recommendation and pilot criteria
- Step 03
Define the landing zone
Design and build the target foundation: identity and access, networking, storage, security policies and observability. Connect the delivery pipeline and validate the setup before moving the pilot workload.
DeliverableLanding zone design and a validated pilot environment
- Step 04
Pilot
Move one representative workload to the target. Check behaviour, data integrity, performance and recovery against the baseline.
DeliverablePilot results and decision on the first wave
- Step 05
Migrate in waves
Group by dependencies and risk. Rehearse the cutover and recovery procedure, then validate each wave before starting the next.
DeliverableWave plan, cutover record and sign-off
- Step 06
Stabilise & hand over
Monitor the workloads on the new platform, transfer runbooks and retire the source only after agreed checks and sign-off.
DeliverableOperations handover and retirement plan
Before each cutover, agree:
- Who makes the go or no-go decision
- How long rollback remains feasible
- How to recover if writes on the target cannot be reversed safely
[ Illustrative cloud-native use case ]
Move a Kubernetes application while both platforms are live
Imagine an order platform on Amazon Elastic Kubernetes Service (EKS). The team wants to bring steady workloads back to private infrastructure. It needs to keep serving customers while the two environments run side by side.
- 01
Choose where each service belongs
For private infrastructure, compare VMware vSphere Kubernetes Service (VKS) and Red Hat OpenShift. If the business case favours another public cloud, evaluate Google Kubernetes Engine (GKE) or Azure Kubernetes Service (AKS). Use the existing EKS setup as the baseline, and keep services there when replacing an AWS dependency costs more than moving would save.
- 02
Design the landing zone for both platforms to coexist
This is often the hardest part. The target has to fit the application and work alongside the source. Map workload identity, secrets, DNS, ingress, network policies and storage before deciding how the clusters connect.
Use CI/CD to build and test one image, then deploy it through GitOps with configuration for each cluster. Where service-to-service traffic needs shared identity and traffic controls, evaluate a full multicluster service mesh with mutual TLS. Test service discovery, latency and failure behaviour before using it in production.
- 03
Pilot traffic and rehearse the data cutover
Pilot a stateless order API and shift a small share of traffic to the target. Check error rates and latency before expanding. For services with persistent data, rehearse replication and restore, verify consistency and agree when writes switch to the target. Once the target accepts writes, recovery may require data reconciliation rather than a simple traffic reversal.
One release, two live clusters
The same application version is deployed to both environments. Customer traffic shifts independently as the target passes its checks.
CI/CD + GitOps
Build and test one image. Deploy it to both clusters with environment-specific configuration.
Weighted ingress gateway
Route requests to the source and target at the agreed split.
Amazon EKS cluster
Current order API remains live during the pilot.
Private Kubernetes cluster
Pilot order API on the chosen platform, with identity, networking, policies and monitoring in place.
Source ↔ Target: optional multicluster service mesh. Mutual TLS and service-to-service traffic policies for calls that cross clusters.
Example traffic progression for the stateless API
Advance only if error rate, latency and application checks pass. Pause or shift API traffic back if they fail. Stateful services follow a separate data cutover and recovery plan.
[ Inside the pilot ]
Make the first move count
Choose a workload that exposes a real migration risk, such as a stateful service, a critical integration or a tight cutover window. Keep the source available where feasible while the target is tested.
What the pilot should tell you
- Which dependencies carry over and which need work
- Whether application behaviour and data meet the agreed checks
- How long the cutover takes and whether recovery meets your requirements
- Which workloads belong in the first wave and what must change before they move
[ Business requirements ]
What to know before you move
Start with the event driving the decision and the conditions a move has to meet. That gives you a way to compare staying, moving selected workloads or migrating the estate.
- Trigger 01
A renewal or cloud commitment is coming up
Licences, hosting or committed spend are due for renewal. You need time to compare terms and test a viable alternative.
What to establishThe decision deadline, current commitments and the full cost of each option, including migration and the period when both platforms run.
- Trigger 02
The cost of running steady workloads is rising
Predictable workloads may cost less elsewhere, while burst capacity or managed services may still make sense on the current platform.
What to establishCost per workload, usable capacity, data transfer charges and the payback period after migration and operating costs.
- Trigger 03
A customer or policy sets new data requirements
New residency, access or audit requirements may change where an application can run or who can operate it.
What to establishWhich data and services are in scope, where they can live and what evidence the security or risk owner needs before sign-off.
- Trigger 04
A platform is losing support or getting harder to run
End of support, duplicate platforms or stretched teams can make consolidation or a move to managed Kubernetes worth evaluating.
What to establishThe support deadline, acceptable downtime and data loss, and who will own the platform after cutover.
Use these requirements to set the pilot’s pass criteria and decide which workloads should move first, move later or stay.
[ Get started ]
Find the first workload worth moving
Start with a migration review. Bring your current environment, your reason for moving and any deadline or target you already have.
- Identify the constraints that could change the decision
- Discuss a candidate workload and what the pilot should prove
- Agree what information is needed to scope the next step

