MLA-C01 Question 244
Select 2You are deploying a machine learning model for real-time predictions using Amazon SageMaker. The model needs to handle high throughput with low latency requirements, and you want to optimize costs while ensuring scalability. Which combination of features should you use?
- A
Enable Amazon SageMaker Model Autoscaling
- B
Deploy the model on a multi-model endpoint
- C
Use Amazon SageMaker Serverless Inference
- D
Deploy the model across multiple Availability Zones
- E
Enable Amazon SageMaker Debugger for performance monitoring
Show answer and explanation
Correct answers: A, B
Explanation
To address high throughput and low latency while optimizing costs, enabling Amazon SageMaker Model Autoscaling ensures that the endpoint scales dynamically based on demand. Deploying the model on a multi-model endpoint allows multiple models to share the same resources, further optimizing costs. Serverless Inference and Debugger do not directly address the scenario requirements, and while deploying across multiple Availability Zones improves availability, it is not specifically required for cost and scalability optimization.
- A. Correct.
Correct: Enabling Amazon SageMaker Model Autoscaling allows the system to handle variable traffic by dynamically adjusting the number of instances, ensuring both scalability and cost optimization.
- B. Correct.
Correct: Deploying the model on a multi-model endpoint allows hosting multiple models on a single endpoint, reducing costs and enabling efficient resource utilization.
- C. Incorrect.
Incorrect: Amazon SageMaker Serverless Inference is suitable for low-latency, intermittent workloads but may not handle high throughput cost-effectively compared to autoscaling solutions.
- D. Incorrect.
Incorrect: Deploying across multiple Availability Zones improves availability but does not directly address cost optimization or scalability for high throughput and low latency.
- E. Incorrect.
Incorrect: Amazon SageMaker Debugger helps in monitoring and debugging training jobs, but it does not assist in optimizing costs or scaling for real-time inference.