MLA-C01 Question 315
Select 3You are managing a real-time machine learning application deployed on an Amazon SageMaker endpoint. The application experiences high traffic during business hours (9 AM to 6 PM) and low traffic during off-peak hours. To optimize the cost while ensuring low latency during high-demand periods, how should you configure SageMaker endpoint auto scaling policies?
- A
Set up a target tracking scaling policy based on the invocation request count per instance to adjust capacity dynamically.
- B
Configure a scheduled scaling policy to increase the number of instances before 9 AM and scale down after 6 PM.
- C
Enable Elastic Load Balancing (ELB) to automatically distribute traffic across multiple SageMaker endpoints.
- D
Use a fixed number of instances in the endpoint configuration to handle peak traffic without scaling.
- E
Set up an alarm-based scaling policy using Amazon CloudWatch metrics such as CPU utilization to trigger scaling activities.
Show answer and explanation
Correct answers: A, B, E
Explanation
To meet scalability requirements while optimizing costs, you should combine target tracking, scheduled scaling, and alarm-based scaling policies. Target tracking scaling adjusts the number of instances dynamically based on metrics like invocation request count. Scheduled scaling handles predictable traffic patterns, such as scaling up before business hours and scaling down afterward. Alarm-based scaling ensures responsiveness to unexpected traffic spikes by monitoring metrics like CPU utilization. These policies together provide a flexible, cost-effective, and scalable solution for managing SageMaker endpoint traffic.
- A. Correct.
Correct. Target tracking scaling policies allow SageMaker to dynamically adjust the number of instances based on a specific metric, such as invocation request count per instance, ensuring scalability during fluctuating demand.
- B. Correct.
Correct. Scheduled scaling policies are ideal for predictable traffic patterns, such as scaling up before 9 AM and scaling down after 6 PM in this scenario.
- C. Incorrect.
Incorrect. Elastic Load Balancing (ELB) distributes traffic but does not directly scale SageMaker endpoints. While ELB can work alongside SageMaker, it is not the appropriate solution for this scenario.
- D. Incorrect.
Incorrect. Using a fixed number of instances is cost-inefficient and does not dynamically adjust to fluctuating demand, potentially leading to underutilization or latency issues.
- E. Correct.
Correct. Alarm-based scaling policies trigger scaling actions based on metrics like CPU utilization, which can help maintain performance during unexpected traffic spikes.