MLA-C01 Question 389
Select 2You are managing a machine learning model hosted on Amazon SageMaker, and you notice a significant increase in the monthly cost of your infrastructure. The model is used for low-latency inference, and traffic patterns indicate that peak usage occurs during specific hours of the day. Which of the following actions should you take to optimize infrastructure and reduce costs while maintaining performance?
- A
Implement Amazon SageMaker endpoint auto-scaling to match the number of deployed instances with traffic patterns.
- B
Switch to an Amazon SageMaker asynchronous inference endpoint to handle requests without maintaining active instances.
- C
Use Spot Instances for your Amazon SageMaker endpoint to reduce costs during off-peak hours.
- D
Monitor endpoint utilization using Amazon CloudWatch and adjust the number of instances manually during peak hours.
- E
Move the model to an Amazon EC2 instance and implement a custom load balancer to handle traffic.
Show answer and explanation
Correct answers: A, D
Explanation
To optimize infrastructure and costs for a low-latency inference use case, it is essential to dynamically scale resources to match traffic patterns. SageMaker endpoint auto-scaling automatically adjusts instance counts based on demand, reducing costs during off-peak hours. Monitoring utilization with Amazon CloudWatch and making manual adjustments can also help fine-tune resource allocation. Other options, such as Spot Instances or asynchronous inference, are either unsuitable for low-latency requirements or introduce unnecessary complexity.
- A. Correct.
Correct: Amazon SageMaker endpoint auto-scaling enables dynamic adjustments to the number of instances based on traffic patterns, ensuring cost optimization without impacting performance.
- B. Incorrect.
Incorrect: Asynchronous inference is not suitable for low-latency requirements as it introduces additional latency due to request queuing and processing.
- C. Incorrect.
Incorrect: Spot Instances are not recommended for low-latency inference because they can be interrupted, which may affect availability and performance.
- D. Correct.
Correct: Monitoring endpoint utilization with Amazon CloudWatch allows you to make manual adjustments during peak hours, ensuring resources are not over-provisioned during off-peak times.
- E. Incorrect.
Incorrect: Moving the model to Amazon EC2 and implementing a custom load balancer would increase management complexity and may not offer the same cost-efficiency as SageMaker's built-in solutions.