MLA-C01 Question 314
Select 3You are managing a machine learning model deployed on an Amazon SageMaker endpoint that experiences fluctuating usage throughout the day. During business hours, the endpoint handles high traffic, while at night, traffic significantly decreases. To optimize cost and maintain performance, which actions should you take to implement a scalable solution?
- A
Configure an Amazon SageMaker endpoint auto-scaling policy based on the number of invocations per instance.
- B
Set up a fixed instance count for the endpoint to maintain consistent performance regardless of traffic.
- C
Use a scheduled scaling policy to adjust the instance count during peak hours and scale down during off-peak hours.
- D
Enable automatic instance replacement to ensure instance health during high traffic periods.
- E
Monitor SageMaker endpoint metrics in Amazon CloudWatch and define custom scaling policies based on demand.
Show answer and explanation
Correct answers: A, C, E
Explanation
To meet scalability requirements and optimize costs for fluctuating traffic, combining auto-scaling policies with metrics-based monitoring and scheduled scaling is an effective strategy. Auto-scaling policies based on invocations per instance dynamically adjust resources in response to demand, while scheduled scaling can handle predictable traffic spikes. Monitoring metrics in CloudWatch provides insights to fine-tune these policies, ensuring the solution remains efficient and responsive.
- A. Correct.
Correct: Configuring an auto-scaling policy based on invocations per instance ensures that the endpoint scales up or down dynamically as traffic fluctuates, optimizing both performance and cost.
- B. Incorrect.
Incorrect: Setting a fixed instance count does not account for fluctuating traffic and may lead to over-provisioning during low traffic or under-provisioning during peak traffic.
- C. Correct.
Correct: Scheduled scaling policies allow you to preemptively adjust instance counts for known traffic patterns, such as scaling up during business hours and scaling down at night.
- D. Incorrect.
Incorrect: Automatic instance replacement is a feature for maintaining instance health, not for handling scalability requirements based on traffic.
- E. Correct.
Correct: Monitoring SageMaker endpoint metrics in CloudWatch allows you to define and refine auto-scaling policies tailored to specific demand patterns, improving responsiveness and cost efficiency.