Free ยท Lesson 1 of 11
DMS copies data from a source endpoint to a target endpoint by running a task. The compute is either a replication instance you size, or a serverless task with a capacity range. Those are different choices, and both still exist.
Sign in to save your place.
The endpoints are the two warehouses. The task is the work order: move everything, keep copying changes, or both. The replication instance is a truck you rent by the hour. Serverless is a truck yard that assigns trucks inside a range you set. You still write the work order.
A source endpoint describes how to log into the database you are leaving. A target endpoint describes the database you are filling. A replication task says which tables move and whether you want a full load, CDC, or both.
A replication instance is an Amazon-managed compute node that runs the task. You pick an instance class and storage. It exists, and it can cost money, even when the task is idle. Multi-AZ is an option on that instance, not a property of every task.
DMS Serverless is a different way to run a task. You set capacity bounds. AWS documents capacity in DMS capacity units. The minimum, the maximum, and what happens when a task is idle have been updated in the service docs. Read the current serverless page before you tell someone it scales to zero or that it replaced replication instances. It has not removed them.
| Piece | You create it | It is not |
|---|---|---|
| Endpoint | Connection details for one database | The task |
| Replication instance | A sized computer for tasks | Required for every serverless task |
| Serverless task | A task plus a capacity range | A schema conversion project |
| Task | Tables, rules, and full load or CDC | A backup of the source |
DMS does not choose your cutover window, your application connection string, or whether a stored procedure was translated. Heterogeneous schema and code conversion is a separate step, covered later. Homogeneous moves can still fail on extensions, large objects, and sequences.
Supported engines change. The user guide's sources and targets pages are the list. A conference slide is not.
Step 1 of 3The source endpoint is a connection, not a copy of the data.
Step 2 of 3The task is the work order. It runs on an instance or as a serverless task.
Step 3 of 3The target endpoint receives rows. It does not, by itself, rewrite application SQL.
Write this for a move you actually care about. Use "unknown" where you have not decided. Unknown is a better answer than a guess.
Source engine and version:
Target engine and version:
Compute: replication instance OR serverless (undecided is allowed):
Task type: full load, CDC, or both:
Who will change the application connection string:
Checked against the linked documentation in October 2026. A later release can move a flag, a price, or a supported pair. Use the page for the version you installed.