Applies toSQL Server 2008–2022 on-premises; Azure SQL Database, Azure SQL Managed Instance, SQL Server on Azure VMs, Amazon RDS for SQL Server, RDS Custom, SQL Server on EC2, Aurora PostgreSQL with Babelfish
1. The 6Rs applied to SQL Server
| R | For SQL Server it means | Typical share of an estate |
|---|---|---|
| Rehost | Same SQL Server on a cloud VM (Azure VM / EC2) | 30–50% |
| Replatform | Managed SQL Server (Azure SQL MI, Azure SQL DB, RDS) | 30–50% |
| Refactor | Change engine (Aurora PostgreSQL / Babelfish, PostgreSQL) | 5–15% |
| Retire | Unused databases, old copies | 10–20% (always more than expected) |
| Retain | Stays on-premises (latency, regulation, end-of-life app) | Some |
| Repurchase | Replace the app with SaaS | Some |
2. Targets side by side
| Target | SQL Server compatibility | You manage | Notes |
|---|---|---|---|
| SQL Server on Azure VM / EC2 | 100% | OS, SQL, patching, HA, backups (helped by the SQL IaaS extension on Azure) | Fastest move; least modernisation |
| Azure SQL Managed Instance | Very high (Agent, cross-DB queries, CLR, linked servers, Service Broker within the instance) | Schema and queries | The usual replatform target for apps that use instance features |
| Azure SQL Database | Database-scoped only (no Agent, no cross-DB queries) | Schema and queries | Best for modern, single-database apps; serverless and Hyperscale options |
| Amazon RDS for SQL Server | High, with restrictions (no sysadmin, limited OS-level features) | Schema and queries | Managed backups, Multi-AZ |
| RDS Custom for SQL Server | Near 100% with OS and sysadmin access | Some OS tasks; AWS automates the rest | For apps needing OS access, custom agents or settings |
| Aurora PostgreSQL + Babelfish | Partial T-SQL and TDS protocol support | Code changes for unsupported features | Removes SQL Server licences; requires testing |
3. Decision flow
Does the app need OS access, unsupported features or a specific SQL version? → VM / EC2 (or RDS Custom on AWS)
Uses SQL Agent, cross-database queries, CLR, linked servers, DB Mail? → Azure SQL MI / RDS (check the feature list)
Single database, modern app, no instance features? → Azure SQL Database (or RDS)
Strong goal to drop SQL Server licences and code can change? → Babelfish / PostgreSQL (refactor)
Cloud choice already made by the business? → Use that cloud's equivalent row
4. Waves
- Wave 0 (pilot): 3–5 low-risk databases, to prove networking, identity, backups and monitoring.
- Wave 1: dev and test copies, plus small Tier 3 apps.
- Waves 2–n: grouped by application, not by server, because apps that share databases or linked servers have to move together.
- Last waves: Tier 1 and complex databases, with rehearsed cutovers.
5. Success measures
- Databases migrated per month, against plan.
- Incidents and rollbacks per wave.
- Performance against the on-premises baseline (p95 query time, batch duration).
- Run cost against the business case, including licence savings.
Tip
Don't pick one target for the whole estate. The best programmes use a mix: managed instances for most apps, VMs for the awkward 20%, and refactoring only where there's a clear payback.
In this series: discovery and assessment, licensing, sizing, cutover and validation, then step-by-step guides for each Azure and AWS path.