MLS-C01 Question 249
Select 2You are deploying a machine learning model using Amazon SageMaker. After deployment, you notice that the latency of the inference requests has increased during certain periods of the day. You suspect this is due to fluctuating traffic. What steps can you take to address this issue while optimizing costs?
- A
Enable Amazon SageMaker endpoint auto-scaling to adjust the number of instances based on traffic patterns.
- B
Switch to a larger instance type for the SageMaker endpoint to handle peak traffic.
- C
Implement asynchronous inference in SageMaker to process requests in batches during high traffic.
- D
Configure Amazon CloudWatch alarms to monitor endpoint metrics and manually adjust instance count when needed.
- E
Use multi-model endpoints in SageMaker to deploy multiple models on the same instance.
Show answer and explanation
Correct answers: A, C
Explanation
To address fluctuating traffic and optimize costs, enabling Amazon SageMaker endpoint auto-scaling ensures that the number of instances dynamically adjusts based on traffic patterns, reducing latency during high traffic and saving costs during low traffic. Additionally, asynchronous inference can decouple processing and request submission, allowing for more efficient handling of bursts in traffic while maintaining cost efficiency. These approaches provide scalable and cost-effective solutions to the problem.
- A. Correct.
Enabling Amazon SageMaker endpoint auto-scaling allows the service to automatically adjust the number of instances based on real-time traffic patterns, which can reduce latency during high traffic periods while optimizing costs.
- B. Incorrect.
Switching to a larger instance type can handle peak traffic, but it may lead to higher costs and underutilization during low traffic hours. This approach is not the most cost-efficient solution.
- C. Correct.
Implementing asynchronous inference processes requests in batches, which can help reduce latency during high traffic by decoupling request submission from processing. This approach is cost-efficient for workloads with sporadic or unpredictable traffic patterns.
- D. Incorrect.
Configuring Amazon CloudWatch alarms and manually adjusting the instance count might address the issue, but it adds operational overhead and is less efficient compared to auto-scaling.
- E. Incorrect.
Multi-model endpoints are used to host multiple models on the same instance, which is beneficial for cost-saving in certain scenarios. However, it does not address fluctuating traffic or latency issues for a single model.