MLS-C01 Question 253
Select 2You are building a machine learning model that will be deployed to predict customer churn for an e-commerce platform. The ML model is expected to handle a high volume of real-time prediction requests globally, with minimal latency. Additionally, the solution must be fault-tolerant and automatically scale during traffic spikes. Which combination of AWS services and features should you use to meet these requirements?
- A
Deploy the model using Amazon SageMaker Hosting with auto-scaling enabled.
- B
Use Amazon CloudFront to distribute requests across global endpoints for low latency.
- C
Host the model on an Amazon EC2 instance with an Elastic IP attached.
- D
Configure Amazon SageMaker multi-model endpoint to serve multiple models simultaneously for fault tolerance.
- E
Use AWS Elastic Load Balancer in front of the SageMaker endpoint to distribute traffic.
Show answer and explanation
Correct answers: A, B
Explanation
To meet the requirements of performance, availability, scalability, and fault tolerance, deploying the model using Amazon SageMaker Hosting with auto-scaling ensures that the solution can handle traffic spikes. Additionally, using Amazon CloudFront reduces latency by serving requests from edge locations closest to the users, improving global availability and responsiveness. Other options, such as using a single EC2 instance or relying on multi-model endpoints, do not fully meet the requirements of the scenario.
- A. Correct.
Correct: Amazon SageMaker Hosting with auto-scaling ensures that the model can handle varying traffic loads by automatically scaling up or down, making it suitable for high availability and scalability.
- B. Correct.
Correct: Amazon CloudFront is a content delivery network that helps reduce latency by delivering requests from edge locations closest to the user, improving the global performance of the ML solution.
- C. Incorrect.
Incorrect: Hosting the model on a single EC2 instance does not provide fault tolerance or scalability. An Elastic IP only ensures a static IP address, but does not address traffic spikes or high availability.
- D. Incorrect.
Incorrect: SageMaker multi-model endpoints are designed to host multiple models on the same endpoint for cost efficiency, but they do not inherently provide fault tolerance for a single model.
- E. Incorrect.
Incorrect: An Elastic Load Balancer cannot directly distribute traffic to a SageMaker endpoint. SageMaker endpoints are managed services that handle their own load balancing.