MLA-C01 Question 317
Select 2You are deploying a machine learning model using Amazon SageMaker, and your model needs to handle fluctuating traffic patterns. During peak hours, the endpoint should scale up to manage high demand, while during off-peak hours, the endpoint should scale down to save costs. Which steps should you take to implement an auto-scaling policy for your SageMaker endpoint?
- A
Create an Amazon CloudWatch alarm to monitor a metric, such as InvocationsPerInstance, and link it to an auto-scaling policy.
- B
Configure the SageMaker endpoint to automatically scale horizontally without any additional configuration.
- C
Set up a target tracking scaling policy in Application Auto Scaling for the SageMaker endpoint.
- D
Manually adjust the number of instances for the SageMaker endpoint based on daily traffic patterns.
- E
Define a desired instance count and associate it with the auto-scaling policy to maintain the required capacity.
Show answer and explanation
Correct answers: A, C
Explanation
To meet scalability requirements and handle fluctuating traffic, you can use SageMaker's integration with Application Auto Scaling. This involves setting up a CloudWatch alarm to monitor metrics, such as InvocationsPerInstance, and configuring a target tracking scaling policy. These mechanisms ensure that the endpoint scales up during high demand and scales down during low demand, optimizing both performance and cost.
- A. Correct.
Correct: Amazon CloudWatch alarms can be used to monitor metrics such as InvocationsPerInstance, which can trigger scaling actions when thresholds are breached.
- B. Incorrect.
Incorrect: SageMaker endpoints do not scale automatically without explicit configuration of auto-scaling policies.
- C. Correct.
Correct: Target tracking scaling policies in Application Auto Scaling allow you to automatically scale SageMaker endpoints based on predefined metrics like average invocations per instance.
- D. Incorrect.
Incorrect: Manually adjusting instance counts is not an automated solution and does not leverage SageMaker's auto-scaling capabilities.
- E. Incorrect.
Incorrect: Defining a desired instance count does not enable auto-scaling; instead, it sets a static number of instances.