MLS-C01 Question 247
Select 2You are deploying a machine learning model on Amazon SageMaker. The model will handle real-time predictions with low latency requirements. During testing, you notice the endpoint is underperforming during peak traffic. Which of the following steps would improve performance while adhering to AWS best practices?
- A
Enable Auto Scaling for the endpoint.
- B
Use Elastic Inference to add GPU acceleration to the endpoint.
- C
Increase the instance size of the endpoint to a larger instance type.
- D
Set up an Amazon SQS queue in front of the endpoint to handle incoming requests.
- E
Use Amazon SageMaker asynchronous inference instead of real-time inference.
Show answer and explanation
Correct answers: A, B
Explanation
To improve performance for a real-time, low latency SageMaker endpoint, enabling Auto Scaling ensures the endpoint can handle fluctuating traffic dynamically, and using Elastic Inference can boost performance for deep learning models in a cost-effective way. These approaches align with AWS best practices for scalable and efficient model deployment.
- A. Correct.
Enabling Auto Scaling allows the endpoint to dynamically adjust the number of instances to handle varying levels of traffic, ensuring consistent performance during peak times.
- B. Correct.
Elastic Inference provides GPU acceleration for certain deep learning models, which can improve inference performance without the cost of a full GPU instance.
- C. Incorrect.
Increasing the instance size could improve performance, but relying solely on larger instances may not efficiently handle fluctuating traffic or be cost-effective.
- D. Incorrect.
Using an Amazon SQS queue could help decouple workloads, but it is not suitable for real-time inference as it introduces latency.
- E. Incorrect.
Asynchronous inference is designed for batch processing or non-real-time use cases, and does not meet the requirement for low latency predictions.