Migration Design and the Cloud Adoption Framework
AZ-305 AZ-305 · Design infrastructure solutions
Migration is where architecture meets reality: moving real workloads, with real dependencies and real downtime constraints, into Azure. AZ-305 tests migration as a structured discipline — the Cloud Adoption Framework, the assessment and discovery phase, the choice of migration strategy per workload, and the tooling for servers, databases, and data.
The Cloud Adoption Framework (CAF)
The Microsoft Cloud Adoption Framework for Azure is the methodology for cloud adoption: it covers strategy, planning, readiness, adoption (migrate, innovate), governance, and management. For AZ-305, the key CAF concepts are the landing zone (the pre-configured environment with networking, identity, governance, and security ready before workloads arrive) and the migration strategies — the "5 Rs": Rehost (lift and shift — move as-is to VMs), Refactor (replatform — move to PaaS with minimal changes, like SQL Server to Azure SQL Managed Instance), Rearchitect (rebuild as cloud-native, such as decomposing a monolith into microservices), Rebuild (discard and rebuild), and Replace (move to SaaS). The exam tests strategy selection: rehost for speed when the workload is stable; refactor when PaaS reduces operational cost with modest effort; rearchitect only when the business case justifies the investment. The landing zone must come first — migrating workloads into an unprepared subscription creates technical debt that is expensive to fix later.
Discovery and assessment: Azure Migrate
Azure Migrate is the hub for discovery, assessment, and migration. It discovers on-premises servers (VMware, Hyper-V, physical), assesses their readiness for Azure (compatibility, sizing, cost estimates), and tracks the migration. The assessment answers the questions that drive the business case: which VMs are ready to migrate, what Azure size they need, and what they will cost. Dependency analysis (via Service Map or the dependency visualisation agent) maps which servers talk to each other, so migration waves group dependent workloads together — migrating a web server without its database is a classic failure. The exam tests this: always assess and map dependencies before migrating, and migrate in waves by dependency group.
Migrating servers: rehost and replatform
For rehost, Azure Migrate's server migration replicates VMs to Azure with minimal downtime (agentless for VMware, agent-based for Hyper-V and physical servers), then a cutover window switches traffic. For replatform, the target depends on the workload: web apps to App Service, containers to AKS or Container Apps, databases to Azure SQL. Azure Migrate's app and code assessment tools analyse .NET and Java applications for cloud readiness. The exam's key judgement: rehost is fastest but preserves the operational burden of VMs; replatform takes longer but reduces long-term cost and management. When a scenario emphasises speed and minimal change, rehost wins; when it emphasises cost reduction and modernisation, replatform wins.
Migrating databases
Azure Database Migration Service (DMS) and the Azure SQL Migration extension for Azure Data Studio handle database migrations with minimal downtime: they replicate data continuously and cut over when ready. The process: assess compatibility (Data Migration Assistant or Azure Migrate), choose the target (Azure SQL Database, Managed Instance, or SQL on VMs), migrate the schema, replicate data, and cut over. For online migrations, DMS uses continuous sync so the source stays live until cutover; for offline migrations, the database is unavailable during the move — acceptable only when downtime is tolerable. The exam tests the choice: online migration when downtime must be minimal; offline when the database is small or downtime is acceptable. Also remember that DMS supports heterogeneous migrations (Oracle or MySQL to Azure SQL) via the Database Migration Service or Azure Data Factory.
Migrating unstructured data
For large data volumes, the network is the bottleneck. AzCopy is the command-line tool for copying blobs, files, and tables — fast, resumable, and scriptable, suitable for online transfers. Azure Data Box is the offline answer: Microsoft ships a ruggedised storage device (up to 80 TB usable), you copy data locally, ship it back, and Microsoft uploads it to Azure. Use Data Box when the data volume makes network transfer impractical (limited bandwidth, tight timeline, or petabytes of data). Azure Data Factory can also move data with transformation. The exam rule of thumb: calculate the transfer time — if the network cannot move the data in the available window, Data Box is the answer. Azure File Sync keeps on-premises file servers synchronised with Azure Files during and after migration, enabling a gradual transition.
Migration waves and cutover
Migrations run in waves: groups of interdependent workloads migrated together, ordered by dependency and business priority. Each wave follows the same pattern: replicate, test in Azure (isolated test failover), schedule cutover, switch DNS and traffic, validate, and decommission the source. Rollback planning is essential: keep the source running until the Azure deployment is validated, and know how to switch back. The exam rewards designs that minimise risk: test migrations before cutover, migrate non-critical workloads first to build confidence, and never decommission the source until the new environment is proven.
Post-migration optimisation
Migration is not finished at cutover. The weeks after migration are when the real value is captured — or lost. Right-size migrated VMs: on-premises VMs are usually oversized, and Azure Migrate's assessments are estimates, so review actual utilisation after 30 days and resize. Modernise in place: a rehosted VM running SQL Server can later move to Azure SQL Managed Instance, and a rehosted web server can later move to App Service — plan these as follow-on waves, not as part of the initial migration. Optimise cost: buy reservations once steady-state usage is known, enable auto-shutdown for non-production, and review storage tiers for migrated data. Decommission the on-premises infrastructure only after the Azure environment has run successfully through a full business cycle (including month-end and any seasonal peaks), and keep backups of the source until decommissioning is complete. Validate the business case: compare actual Azure costs against the pre-migration estimate and report the variance — this builds credibility for the next migration wave and surfaces optimisation opportunities.
Planning downtime and cutover windows
Every migration needs a cutover plan, and the plan depends on how much downtime the business tolerates. For zero-downtime migrations, use continuous replication (Azure Migrate server migration or DMS online migration) and cut over by switching DNS or connection strings in a short window, with the source kept running for rollback. For planned-downtime migrations, schedule the cutover in a maintenance window, stop writes to the source, perform a final sync, and switch over — simpler and cheaper, but the window must be long enough for the final sync. Always define the rollback criteria in advance: what failure would trigger a rollback, who decides, and how long the team has to decide before the window closes. Test the cutover in a non-production wave first. The exam tests this judgement: a scenario with a strict downtime limit points to online migration with replication; a scenario with a generous weekend window points to the simpler offline approach.
Exam traps to watch for
- Migrating workloads before the landing zone is ready — governance, networking, and identity must exist first.
- Migrating a server without its dependencies — always map dependencies and migrate in waves by dependency group.
- Choosing online database migration when the scenario allows downtime — offline is simpler and cheaper when downtime is acceptable.
- Using network transfer for petabytes of data on a slow link — calculate transfer time; Data Box exists for exactly this case.
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.
Q1.A company is starting its Azure migration. The platform team wants to establish networking, identity, governance, and security before any workload moves. What should the architect recommend they build first?
Explanation: A landing zone is the pre-configured environment (networking, identity, governance, security) that workloads migrate into — the CAF's foundational step. Option B is wrong because a scale set is a workload component, not a foundation. Option A is wrong because Data Factory moves data; it does not establish the environment. Option C is wrong because per-workload VPNs are connectivity details, not the platform foundation.
Exam objective: AZ-305 Domain 4: Design migrations
Q2.A company must migrate 500 VMs to Azure as quickly as possible with minimal changes. The applications are stable and the team has limited time. Which migration strategy should the architect recommend?
Explanation: Rehost moves VMs as-is with minimal change and the fastest timeline — the right strategy when speed matters and the applications are stable. Option D is wrong because rearchitecting 500 VMs' applications would take far longer. Option A is wrong because replacing applications with SaaS is a different project with its own evaluation. Option C is wrong because rebuilding is the slowest and most expensive strategy.
Exam objective: AZ-305 Domain 4: Design migrations
Q3.Before migrating, a company needs to know which on-premises servers communicate with each other so that dependent workloads move together. Which Azure Migrate capability provides this?
Explanation: Dependency analysis maps network connections between servers, showing which workloads depend on each other so migration waves can group them. Option B is wrong because assessment checks readiness and sizing, not dependencies. Option A is wrong because cost estimation projects Azure spend. Option D is wrong because business case analysis builds the financial justification.
Exam objective: AZ-305 Domain 4: Design migrations
Q4.A company must migrate a 2 TB production SQL Server database to Azure SQL Managed Instance with less than 30 minutes of downtime. Which approach should the architect recommend?
Explanation: An online DMS migration continuously replicates changes while the source stays live, then cuts over in minutes — meeting the 30-minute downtime requirement. Option A is wrong because an offline migration would leave the database unavailable for hours while 2 TB transfers. Option C is wrong because BACPAC export/import is offline and slow for a 2 TB database. Option B is wrong because AzCopy copies blobs and files, not live database contents.
Exam objective: AZ-305 Domain 4: Design migrations
Q5.A company needs to move 400 TB of archival data to Azure. Its internet connection is 100 Mbps and the migration window is two weeks. What should the architect recommend?
Explanation: 400 TB over 100 Mbps would take over a year; Data Box ships ruggedised storage devices for offline transfer, completing the move within the window. Option A is wrong because the network cannot move 400 TB in two weeks. Option D is wrong because Data Factory still transfers over the same slow network. Option C is wrong because File Sync synchronises file shares continuously; it does not solve the bulk transfer problem.
Exam objective: AZ-305 Domain 4: Design migrations
Q6.A company is migrating a three-tier application (web, app, database) to Azure. The database tier must move in the same wave as the app tier because of tight coupling. How should the architect plan the migration?
Explanation: Dependency analysis shows the app and database tiers are tightly coupled, so they must move together in one wave; the web tier can follow after the coupled tiers are validated. Option C is wrong because separating coupled tiers breaks the application. Option B is wrong because migrating the database last leaves the app tier without its data. Option A is wrong because migrating everything at once with no testing maximises risk.
Exam objective: AZ-305 Domain 4: Design migrations
Q7.A company runs an on-premises SQL Server database that uses features supported by Azure SQL Database. The team wants to reduce operational overhead and is willing to make minor application changes. Which strategy and target should the architect recommend?
Explanation: Refactoring to Azure SQL Database (PaaS) removes patching, backup, and HA management while needing only minor changes — the right trade-off for this scenario. Option B is wrong because rehosting to VMs preserves the operational burden the team wants to reduce. Option D is wrong because Cosmos DB is a NoSQL database requiring a data-model rewrite, not minor changes. Option C is wrong because replacing the database with SaaS is a different product decision, not a migration strategy for this workload.
Exam objective: AZ-305 Domain 4: Design migrations
Q8.During a migration, a company wants to keep its on-premises file server synchronised with Azure Files so users can access the same files from both locations during the transition. Which service should the architect recommend?
Explanation: Azure File Sync keeps on-premises file servers synchronised with Azure Files, with cloud tiering to cache frequently accessed files locally — designed for exactly this hybrid transition. Option A is wrong because Data Box is a one-time offline transfer, not ongoing synchronisation. Option D is wrong because scheduled AzCopy is a manual workaround without conflict handling or tiering. Option C is wrong because Site Recovery replicates VMs for disaster recovery, not file synchronisation.
Exam objective: AZ-305 Domain 4: Design migrations
Q9.A company is migrating VMs to Azure and wants to validate that the migrated VMs work correctly before switching production traffic. What should the architect recommend?
Explanation: A test migration runs the replicated VMs in an isolated network for validation without affecting production, so issues are found before cutover. Option C is wrong because cutting over without validation risks a failed migration. Option D is wrong because deleting the source removes the rollback option. Option B is wrong because skipping testing to save time is the highest-risk approach.
Exam objective: AZ-305 Domain 4: Design migrations