MLA-C01 Question 369
Single answerYou are a machine learning engineer tasked with deploying a model for real-time fraud detection using Amazon SageMaker. The model inference requires low latency, and the traffic to the endpoint is expected to vary significantly throughout the day. Which configuration should you choose to ensure cost-efficiency and performance?
- A
Deploy the model on an Amazon SageMaker real-time endpoint with auto-scaling enabled based on invocation metrics.
- B
Deploy the model on an Amazon SageMaker batch transform job and schedule it to run every hour.
- C
Deploy the model using Amazon SageMaker Asynchronous Inference to handle high traffic variability.
- D
Deploy the model as an Amazon Lambda function integrated with an API Gateway for real-time inference.
Show answer and explanation
Correct answer: A
Explanation
Amazon SageMaker real-time endpoints with auto-scaling are the best choice for real-time inference with variable traffic. They allow you to maintain low latency while automatically adjusting the number of instances based on traffic patterns, ensuring cost-efficiency and performance. Other options like batch transform or asynchronous inference are either not suitable for real-time use cases or do not handle traffic variability effectively.
- A. Correct.
This is the correct answer. Amazon SageMaker real-time endpoints with auto-scaling enabled are designed for low-latency use cases with variable traffic. Auto-scaling adjusts the number of instances based on invocation metrics, ensuring cost-efficiency and performance.
- B. Incorrect.
Batch transform is not suitable for real-time use cases as it processes data in batches and is not designed for low-latency inference.
- C. Incorrect.
Amazon SageMaker Asynchronous Inference is designed for cases where responses can tolerate higher latency, which is not appropriate for real-time fraud detection.
- D. Incorrect.
While AWS Lambda can provide low-latency inference, it is not suitable for large ML models that require significant memory or compute resources, and it does not inherently handle traffic variability as efficiently as SageMaker real-time endpoints with auto-scaling.