Insights · Modernization

The 6Rs of cloud migration: pick a disposition per workload

AWSIBM Cloud

Key takeaway

Migration is decided workload by workload: rehost, replatform, refactor, repurchase, retire or retain, each with a recorded metric.

The framework has a traceable lineage, which matters because the industry keeps arguing about the letter count. Gartner outlined the original 5 R’s in 2011; Stephen Orban’s 2016 AWS post — “6 Strategies for Migrating Applications to the Cloud” — credits it directly and adds repurchase; AWS Prescriptive Guidance now ships seven strategies (Relocate joined); Microsoft’s Cloud Adoption Framework rationalizes the estate into eight dispositions (Rebuild and Replace split out, Rearchitect separated from Refactor); Google publishes its own labels (“lift and shift”, “lift and optimize”, “move and improve”). The count is vendor marketing; the portfolio logic is the framework.

The dispositions, defined

  • Rehost — “lift and shift”: move the workload substantially as-is. Fastest to value, smallest to plan, and the entry ticket to landing-zone maturity. Also the floor of the risk curve: nothing gets fixed in transit.
  • Replatform — “lift, tinker, and shift”: small substitutions en route — managed database instead of self-run, containers instead of VMs — without changing the application’s core architecture.
  • Refactor / re-architect — “re-imagining how the application is architected”: the monolith becomes services, the schema is decomposed, the strangler fig runs. Highest cost, highest ceiling, and — critically — optional.
  • Repurchase — “drop and shop”: leave the product for a SaaS equivalent. The build-versus-buy decision, revisited with cloud leverage.
  • Retire — decommission. AWS’s large-migration guidance identifies “zombie applications” (under 5% CPU/memory) and idle ones (5–20% over 90 days) as retirement candidates; in most portfolios this is the single largest source of savings and the least examined.
  • Retain — keep it where it is, on purpose: compliance pins it, a dependency blocks it, or hardware amortization argues for patience. A deliberate retain is a decision; an accidental one is debt.

(AWS’s seventh, Relocate, is the hypervisor-level transfer of VM estates — “the quickest way to migrate” at platform scale.)

The portfolio discipline: per workload, with metrics

The failure mode of migration programs is uniformity — everything rehosted, or everything refactored. The corrective is a per-workload disposition chosen by value and risk, with the vendor guidance surprisingly aligned:

  • AWS argues against big-bang modernization: “Refactor is not recommended for large migrations … we recommend rehosting, relocating, or replatforming … and then modernizing the application after the migration is complete.” Land, stabilize, then re-architect the workloads whose economics earn it.
  • Microsoft adds decision discipline: business drivers and key indicators mapped to each strategy, per-strategy success metrics (rehost: “migrate 100% of Tier 1 workloads with no SLA degradation”), and the pointed rule “don’t rehost problematic workloads” — a broken application in the datacenter is a broken application in the cloud, with worse latency for the complaint.
  • The academic record explains why discipline matters: the empirical study of legacy-migration challenges (Fahmideh et al., validated with 104 domain experts) ties failures to poor understanding of the migration process itself; the follow-up survey and process-model papers document a “rich inventory” of activities and concerns — and a field whose fragmented terminology is exactly what the Rs-frameworks paper over. Migration is “difficult and high cost” in the literature; only the process discipline bends that curve.

The decision that earns an ADR

Each disposition is a standing architectural commitment, which makes it ADR material: workload tier, chosen R, the rejection reasons (why not refactor? why not retain?), the success metric, and the revisit date. On the E-Tax modernization the dispositions were explicit per component — the core filing estate re-architected via traffic-splitting coexistence, supporting systems rehosted then replatformed, sundry legacy utilities retired outright — and the recorded decisions are what kept the programme from drifting into uniformity under schedule pressure.

Sequencing the estate

  1. Assess and tier. Inventory workloads; score by business criticality, change frequency, and technical health; flag the zombies and idle workloads first — retiring them shrinks everything downstream.
  2. Land the foundation. Landing zone, identity, networking, and infrastructure as code before the first workload moves; the landing zone is the dispositions’ operating environment.
  3. Move the low-risk/high-certainty work first. Rehosts and replatforms build migration muscle while the refactors are being designed, not the reverse.
  4. Modernize deliberately, workload by workload, where the refactor/re-architect case closes — with zero-downtime data discipline for everything that touches state.
  5. Re-rationalize annually. A disposition expires: the rehost of 2026 is the replatform candidate of 2028, and the retire list never stays empty.
Rule of thumb

If every workload in your migration plan has the same R, you don’t have a strategy — you have a habit. The portfolio wins because the dispositions differ.

Rationalize your migration portfolio

Sources