DJames617 Service Comparison
Batch Transform vs Real-Time Inference
The inference pattern follows latency and traffic shape.
What They Have in Common
Each option can solve part of the scenario. The exam expects you to choose the one that satisfies the stated constraints with the right operational burden, security boundary, availability model, and cost behavior.
Key Differences
| Option | Primary model | Best use | Security | Availability | Operations and cost |
|---|---|---|---|---|---|
| Batch transform | Offline scoring | Large scheduled jobs, no persistent endpoint | IAM, S3/KMS input and output controls | Job retries and output durability depend on the workflow | Pay for batch compute; shut down capacity after jobs |
| Real-time endpoint | Synchronous low-latency inference | User-facing predictions, steady or bursty live requests | IAM, VPC, KMS, endpoint and model controls | Deploy multiple instances/AZs for endpoint resilience | Pay for provisioned endpoint capacity; manage scaling and versions |
| Async inference | Large payload or longer inference | Queue-style inference without immediate response | IAM, S3/KMS, endpoint and invocation controls | Managed queueing; design retries and failure handling | Pay for inference and storage; lower always-on pressure than real time |
Typical Exam Clues
Practice After Studying
Return to the certification guide that includes this comparison and launch domain training from there.