Cloud Migration Strategies
The 7 Rs, discovery and portfolio assessment, migration waves, data migration techniques, cutover patterns and the organizational side of large migrations.
The 7 Rs
| Strategy | What it means | Use when |
|---|---|---|
| Retire | Switch off | No business value; duplicates |
| Retain | Keep on-prem for now | Recently upgraded, compliance, low value to move |
| Rehost (lift and shift) | Move VMs as-is | Fast data-center exit; minimal change |
| Relocate | Move hypervisor-level (e.g. VMware Cloud on AWS, Azure VMware Solution, Google Cloud VMware Engine) | Large VMware estates, fastest |
| Repurchase | Move to SaaS | Commodity apps (CRM, HR, email) |
| Replatform (lift, tinker, shift) | Small optimizations: managed DB, containers | Gain managed-service benefits with limited change |
| Refactor/re-architect | Redesign cloud-native | High business value, scale or agility needs |
(The list originated from Gartner's 5 Rs and was extended by AWS.)
Migration process
flowchart LR A[Assess: discovery,<br/>dependency mapping, TCO] --> M[Mobilize: landing zone,<br/>skills, pilot migrations] M --> W[Migrate in waves:<br/>grouped by dependencies] W --> O[Operate & optimize:<br/>rightsizing, modernization]
- Discovery: inventory servers, applications, dependencies (network flow analysis), licences, data volumes; tools like AWS Application Discovery Service, Azure Migrate, Google Migration Center.
- Wave planning: group applications that talk to each other heavily; start with low-risk ones to build the factory.
- Data migration: online replication (AWS DMS, Azure Database Migration Service, Google Database Migration Service, native replication), offline bulk transfer (Snowball, Data Box, Transfer Appliance) for very large datasets.
Cutover patterns
| Pattern | Downtime | Risk |
|---|---|---|
| Big bang with downtime window | Hours | Simple but all-or-nothing |
| Replication + short cutover | Minutes | Common for databases |
| Parallel run with data sync | Near zero | Complex; reconciliation needed |
| Strangler (incremental routing) | None | Best for refactoring |
Architect-level considerations
[!CTO] Rehosting alone rarely saves money; savings come from rightsizing, elasticity, managed services and closing data centers. Budget for a period of double running costs and for skills.
[!ARCHITECT] Define rollback criteria and procedures for every cutover; test them.
Interview questions and answers
A company must exit its data center in 12 months with 800 applications. What is your approach?
Assess and classify quickly (retire/retain/rehost/replatform/refactor), build the landing zone and a migration factory, rehost or relocate the bulk to meet the deadline, replatform where cheap (managed databases), and schedule refactoring of high-value apps after the exit. Govern by waves with clear metrics.
What commonly goes wrong?
Unknown dependencies (hard-coded IPs, shared databases, batch jobs), licensing surprises, underestimating data transfer time, latency between migrated and non-migrated components, and no operating model for the cloud.