Study guide
Technical reference and lesson notes
Purpose of This Lesson
Migration architecture questions often test whether you can match the migration objective to the correct AWS service. The key decisions include:
- Discovering existing server utilization for right-sizing.
- Migrating VMware, physical, and virtual servers with application dependencies.
- Moving and transforming relational database workloads.
- Selecting online transfer versus offline appliances.
- Reducing downtime during server cutover.
- Improving transfer performance when moving large numbers of files.
This lesson focuses on practical patterns using AWS Application Discovery Service, AWS Application Migration Service (MGN), AWS Database Migration Service (DMS), AWS Schema Conversion Tool (SCT), and AWS Snowball Edge.
Key Concepts
AWS Application Discovery Service for assessment and right-sizing
AWS Application Discovery Service collects information about on-premises servers and their workloads before migration. The data can include:
- CPU, memory, and disk utilization.
- Network activity.
- Running processes.
- Installed applications.
- Server configuration details.
- Communication and dependency information, depending on the discovery method.
For VMware environments, the Discovery Connector can be deployed in VMware vCenter. This provides agentless discovery data for virtual machines and is useful when the primary requirement is to assess utilization and determine appropriate AWS instance sizes.
Application Discovery Service is an assessment and planning tool. It does not itself perform the complete server migration.
AWS Application Migration Service for server lift-and-shift
AWS Application Migration Service, commonly called MGN, is designed to replicate and migrate servers into AWS. It supports migration scenarios involving:
- VMware virtual machines.
- Physical servers.
- Other supported virtual or cloud-hosted servers.
- Linux and Windows workloads.
MGN installs a replication agent on supported source servers. The agent continuously replicates data to AWS staging resources. After replication is ready, the server can be launched as an EC2 instance for testing and later cut over to production.
Application groups help organize servers that belong to the same application or migration wave. This is especially important when dependencies exist between web, application, and database tiers. Instead of migrating unrelated servers independently, engineers can group dependent workloads and coordinate testing and cutover.
A typical MGN process is:
- Install the replication agent on source servers.
- Begin continuous replication to AWS.
- Group related servers into application or migration waves.
- Perform test launches.
- Validate application behavior and dependencies.
- Run a final synchronization before cutover.
- Launch production instances and redirect users or traffic.
AWS DMS and AWS SCT for database migration
Migrating a database usually involves two separate concerns:
- Schema and code conversion: handled by the AWS Schema Conversion Tool, or SCT.
- Data replication and movement: handled by AWS Database Migration Service, or DMS.
For example, migrating an Oracle data warehouse to Amazon Redshift may require SCT to assess and convert incompatible schema objects, SQL, procedures, and other database structures. DMS can then migrate the data and support ongoing replication during the transition.
DMS is not a universal automated converter for every database feature. When source and target engines differ, conversion effort and unsupported objects must be assessed. SCT helps identify and address those differences.
Snowball Edge for offline data transfer
AWS Snowball Edge is appropriate when moving data over the network would take too long, be too expensive, or fail to meet the migration deadline. The device is shipped to the customer, data is copied locally, and the device is returned to AWS for import.
A Snowball-based migration is particularly useful when:
- The data volume is large.
- Internet bandwidth is limited.
- The transfer is urgent.
- The source environment cannot sustain a long network transfer.
- Security or operational constraints make online transfer impractical.
For millions of small files, transfer performance can be limited by file-system operations and sequential processing rather than by the device’s raw storage capacity. Parallelizing copy operations can improve throughput. Separate terminals or client processes can be used to run multiple copy jobs concurrently, provided the workload and source storage can support the additional parallelism.
Final synchronization and low-downtime cutover
For servers migrated with MGN, continuous replication keeps the AWS copy close to the source. Before production cutover, perform a final synchronization so that the target contains the latest source changes.
The final cutover pattern is:
- Stop or quiesce application writes on the source.
- Allow the final changes to replicate.
- Confirm replication health and data currency.
- Launch or promote the target environment.
- Redirect users, DNS, or application traffic.
- Validate the application.
The outage is reduced to the time needed to stop writes, complete final synchronization, launch or configure the target, and redirect traffic. Continuous replication does not eliminate the need for a cutover window or application validation.
Exam-Relevant Takeaways
- Use the AWS Application Discovery Service Discovery Connector in VMware vCenter when you need utilization and configuration data for migration assessment and right-sizing.
- Use AWS MGN to migrate supported Linux and Windows servers, including VMware VMs and physical servers, into AWS.
- Use MGN application groups or migration waves when application components have dependencies and must be migrated and tested together.
- Use AWS SCT for schema and code conversion between heterogeneous database engines.
- Use AWS DMS for data movement and ongoing replication during a database migration.
- Use Snowball Edge when a large dataset must be transferred urgently but the available network link is too slow.
- For millions of small files on Snowball Edge, parallel copy operations may improve performance.
- Use MGN’s continuous replication and final synchronization to minimize server migration downtime.
- Discovery, migration, database conversion, database replication, and offline transfer are separate capabilities. Do not select one service merely because it is associated with the same overall migration project.
Architecture Decision Guide
| Requirement | Recommended service or pattern | Why |
|---|---|---|
| Collect VMware VM utilization data for right-sizing | AWS Application Discovery Service Discovery Connector | Gathers environment and performance information from VMware vCenter without performing the migration itself |
| Migrate VMware and physical servers to EC2 | AWS Application Migration Service (MGN) | Provides agent-based continuous replication and controlled cutover |
| Migrate dependent servers in coordinated waves | MGN application groups and migration waves | Keeps related application components organized for testing and cutover |
| Convert Oracle database structures for Amazon Redshift | AWS Schema Conversion Tool | Assesses and converts schema and database code where possible |
| Replicate database data to the target | AWS Database Migration Service | Performs data migration and can maintain ongoing replication during cutover |
| Move 50 TB over a constrained 1 Gbps connection under an urgent deadline | AWS Snowball Edge | Uses physical shipment instead of relying solely on the network link |
| Improve transfer performance for millions of small files | Parallelize Snowball copy operations | Reduces the impact of serial file operations, subject to source and client capacity |
| Minimize downtime for a server migration | MGN continuous replication plus final synchronization | Keeps the target current and limits the outage to the final cutover process |
Common Exam Traps
- Choosing MGN for right-sizing: MGN migrates servers; Application Discovery Service collects assessment data for planning and sizing.
- Confusing DMS with SCT: DMS moves data. SCT converts or assesses schema and database code. A heterogeneous migration may need both.
- Assuming DMS automatically converts all Oracle features to Redshift: Unsupported objects and SQL differences still require assessment and remediation.
- Using network transfer automatically for large data volumes: Calculate transfer time and consider bandwidth, utilization, and the deadline. A physical appliance may be the better choice.
- Assuming Snowball solves all performance problems automatically: Millions of small files can create file-operation overhead. Copy strategy and parallelism matter.
- Ignoring application dependencies: Migrating servers individually can break applications if tiers are cut over in the wrong order or at different times.
- Treating replication as zero downtime: Continuous replication reduces downtime, but writes still need to be controlled and a final synchronization and cutover are still required.
- Assuming discovery tools perform migration: Discovery output supports planning; it does not replace MGN, DMS, or another migration mechanism.
Real-World Engineer Notes
Estimate network transfer time before selecting a service
A 1 Gbps link has a theoretical maximum of approximately 125 MB/s. Transferring 50 TB under ideal conditions would take several days, and actual time will be longer because of protocol overhead, link utilization, throttling, retransmissions, and competing workloads. If the business deadline is shorter than the practical network transfer time, evaluate Snowball Edge.
Validate dependencies before migration waves
Dependency mapping should inform wave design. Grouping servers is not enough by itself; the team should also validate:
- Which services must start together.
- Which systems require low-latency communication.
- Which DNS names or IP addresses are embedded in configuration.
- Whether shared storage or licensing systems are dependencies.
- How traffic will be redirected after cutover.
Test before production cutover
A successful replication state does not prove that the application works in AWS. Perform test launches and validate operating-system configuration, network access, security groups, IAM integration, storage, licensing, monitoring, and application-level behavior.
Treat database conversion as a workload-specific activity
SCT can automate portions of conversion, but database warehouses often contain custom SQL, stored procedures, extensions, scheduling logic, and performance assumptions that do not translate directly. Plan for assessment, remediation, data validation, and performance testing.
Parallel transfers require controlled testing
Running multiple copy operations can improve Snowball throughput, but excessive concurrency can overload the source file system, network, or client host. Increase parallelism gradually and measure throughput, CPU, memory, and error rates.
Quick Reference Summary
- Assess and right-size: Application Discovery Service.
- Discover VMware environment: Discovery Connector in vCenter.
- Migrate servers: AWS MGN.
- Coordinate dependent server migrations: MGN application groups and waves.
- Convert database schema/code: AWS SCT.
- Move and replicate database data: AWS DMS.
- Transfer large datasets over limited bandwidth: Snowball Edge.
- Improve many-small-file transfers: Use parallel copy operations where supported and validated.
- Reduce server cutover downtime: Continuous replication followed by final synchronization.
Flashcards
1. Which AWS service collects utilization data for migration right-sizing?
Answer: AWS Application Discovery Service.
2. What component can collect discovery data from a VMware vCenter environment?
Answer: The AWS Application Discovery Service Discovery Connector.
3. Which AWS service is designed to migrate physical and virtual servers to AWS?
Answer: AWS Application Migration Service, or MGN.
4. Why use MGN application groups?
Answer: To organize servers that belong to the same application or have dependencies and should be migrated, tested, and cut over together.
5. Which service converts database schemas and code between different database engines?
Answer: AWS Schema Conversion Tool, or SCT.
6. Which service performs ongoing database replication during migration?
Answer: AWS Database Migration Service, or DMS.
7. What is the purpose of final synchronization in an MGN migration?
Answer: To replicate the last source changes before launching or promoting the target during cutover.
8. When is Snowball Edge preferable to an online transfer?
Answer: When the dataset is large, network bandwidth is limited, and the required transfer deadline cannot be met reliably over the network.
9. Why can millions of small files transfer slowly to Snowball Edge?
Answer: File-system metadata and per-file operation overhead can limit throughput, even when the device has substantial storage and bandwidth capacity.
10. Does AWS DMS automatically convert all Oracle database code for Amazon Redshift?
Answer: No. SCT is used to assess and convert supported schema and code, while unsupported items may require manual remediation.
Practice Questions
Question 1
A company wants to migrate several hundred VMware virtual machines to AWS. Before selecting EC2 instance types, it must collect several weeks of CPU, memory, disk, and network utilization data from the existing environment. Which solution best meets the requirement?
A. Install the AWS MGN agent on every VM and immediately launch test instances.
B. Deploy the AWS Application Discovery Service Discovery Connector in VMware vCenter.
C. Configure AWS DMS to collect server utilization data.
D. Order a Snowball Edge device and copy the VM disks to it.
Correct answer: B
Explanation: The requirement is discovery and right-sizing, not server migration. The Discovery Connector integrates with VMware vCenter and collects environment data for migration planning. MGN is used later to replicate and migrate the servers.
Question 2
An organization has VMware VMs and physical Linux servers that together form a three-tier application. The web, application, and database servers have dependencies and must be migrated in coordinated waves with minimal downtime. Which approach is most appropriate?
A. Copy each server independently to Amazon S3 and create AMIs manually.
B. Use AWS DMS for all servers and migrate them in alphabetical order.
C. Use AWS MGN, organize related servers into application groups, and perform test launches followed by final synchronization.
D. Use AWS SCT to convert the operating systems before launching EC2 instances.
Correct answer: C
Explanation: MGN is designed for server migration across supported physical and virtual environments. Application groups and migration waves help coordinate dependent components, while continuous replication and final synchronization reduce the cutover outage.
Question 3
A company is migrating an Oracle data warehouse to Amazon Redshift. The source contains custom SQL and database objects that may not be compatible with the target engine. Which solution should the architect recommend?
A. Use only AWS DMS because it converts all schema and code automatically.
B. Use AWS SCT to assess and convert the schema and code, and use AWS DMS to migrate and replicate the data.
C. Use AWS MGN to replicate the Oracle database files directly into Redshift.
D. Use Snowball Edge to convert the Oracle schema during physical transfer.
Correct answer: B
Explanation: SCT addresses heterogeneous schema and code conversion. DMS handles data movement and ongoing replication. The combination supports assessment, conversion, and controlled database cutover.
Question 4
A business must move 50 TB of data to AWS immediately. Its only internet connection is 1 Gbps and is also used by production applications. Which option best satisfies the requirement?
A. Use a single-threaded network copy and accept the transfer time.
B. Use AWS MGN because it is optimized for all bulk file transfers.
C. Use AWS Snowball Edge to perform an offline transfer.
D. Use AWS SCT to compress and transfer the files.
Correct answer: C
Explanation: A 1 Gbps link has a theoretical maximum of about 125 MB/s, and production traffic and protocol overhead reduce the practical rate further. Snowball Edge avoids relying solely on the constrained network and is appropriate for urgent large-scale transfers.
Question 5
A migration team is copying millions of small files to a Snowball Edge device, but the transfer is much slower than expected. Which action is most likely to improve throughput?
A. Run multiple copy operations concurrently from separate terminals or client processes, while monitoring source and client capacity.
B. Convert the files with AWS SCT before copying them.
C. Disable all integrity checks permanently.
D. Move the files with AWS DMS instead.
Correct answer: A
Explanation: Many-small-file workloads are often limited by per-file operation overhead. Parallel copy operations can increase throughput, although concurrency must be tested so it does not overload the source storage, network, or client system.