Serverless DMS tasks remove the need to manage replication instances while still supporting full-load + CDC migrations.
** CDC stands for Change Data Capture **
It’s a mechanism to track and capture changes (inserts, updates, deletes) in a database so that they can be replicated or processed elsewhere.
Think of CDC like a live news feed of database changes. Once you have the full story (full load) + CDC keeps sending you all the updates as they happen.
| Feature | Serverless DMS | Traditional DMS |
|---|---|---|
| Replication instance | ❌ managed by AWS | ✅ provisioned by user |
| Scalability | Auto-scaled | Manual scaling required |
| Cost | Pay per use | Pay hourly for instance |
| Maintenance | Minimal | User responsible for patching, sizing |
| Use case | Sporadic migration, cloud-native, low ops | Continuous heavy workloads, fine-tuned control |
You cannot use serverless DMS in certain situations where the limitations of the service prevent it from working. Here’s a detailed breakdown:
1. Unsupported Database Engines
Serverless DMS does not support all source and target engines.
| Category | Supported? |
|---|---|
| Aurora MySQL | ✅ |
| RDS MySQL | ✅ |
| Aurora PostgreSQL | ✅ |
| RDS PostgreSQL | ✅ |
| Oracle | ❌ (serverless not available) |
| SQL Server | ❌ |
| SAP, Db2, MariaDB | ❌ |
If your migration uses unsupported engines, you must use a provisioned DMS replication instance.
2. Very Large Workloads
- Serverless DMS has limits on database size and throughput.
- For very large databases (e.g., hundreds of TBs) or high CDC volume, a provisioned replication instance may be needed for predictable performance.
3. Advanced Configurations
Serverless DMS cannot handle some advanced replication features, such as:
- Multi-AZ replication configuration (for high availability).
- Fine-grained control over replication instance type, storage, and scaling.
- Large-scale transformations that require custom pre/post-processing tasks.
4. On-Premises Databases Without AWS Direct Connectivity
- If the source database cannot connect reliably to AWS over the network (VPN, Direct Connect, or public endpoints), serverless DMS may fail.
- Provisioned DMS allows more tuning options for network throughput and retry strategies.
Summary Table
| Limitation | When Serverless DMS Won’t Work | Alternative |
|---|---|---|
| Database engine unsupported | Oracle, SQL Server, MariaDB | Use provisioned DMS instance |
| Large-scale workload | Very large databases, high CDC | Provisioned instance |
| Advanced replication features | Multi-AZ, fine-grained control | Provisioned instance |
| Connectivity limitations | On-prem without stable network | Provisioned instance with VPN/Direct Connect |
✅ Key Takeaway:
Serverless DMS is best for small-to-medium workloads, cloud-native databases, and low operational overhead, but for unsupported engines, very large databases, or advanced configurations, you must use a provisioned replication instance.
If you want, I can make an easy-to-remember diagram comparing Serverless DMS vs Provisioned DMS and when to use each, which is very exam-friendly.
Do you want me to do that?
Top comments (0)