MLA-C01 Question 240
Single answerYou are tasked with deploying a machine learning model for a real-time fraud detection system. The model must deliver predictions with low latency and handle up to 10,000 requests per second during peak times. Additionally, the system should scale automatically based on traffic and minimize operational overhead. Which deployment infrastructure would be the most suitable based on these requirements?
- A
Deploy the model as a SageMaker Endpoint with an Auto Scaling policy
- B
Host the model on an EC2 instance with an Elastic Load Balancer
- C
Package the model as a Lambda function and invoke it through API Gateway
- D
Deploy the model to an Amazon ECS cluster using Fargate
Show answer and explanation
Correct answer: A
Explanation
SageMaker Endpoints are designed for real-time inference workloads, offering low latency and built-in auto-scaling capabilities. These features align perfectly with the requirements of handling high request volumes and maintaining low operational overhead. Other options either involve more manual management or fail to meet the latency and scaling requirements effectively.
- A. Correct.
Correct. SageMaker Endpoints are optimized for low-latency, high-throughput real-time inference and support auto-scaling, making them ideal for this use case.
- B. Incorrect.
Incorrect. While EC2 with an Elastic Load Balancer can handle high traffic, it requires significant operational effort to manage scaling, monitoring, and infrastructure maintenance.
- C. Incorrect.
Incorrect. Lambda functions are not ideal for low-latency, high-throughput use cases due to cold start issues and concurrency limits, which may lead to increased latency.
- D. Incorrect.
Incorrect. ECS with Fargate provides container orchestration but lacks the out-of-the-box support for real-time inference that SageMaker Endpoints offer. It would also require additional configuration and management to achieve low latency and auto-scaling.