MLS-C01 Question 251
Select 3A company is deploying a machine learning model in production to handle real-time predictions. The model needs to process thousands of requests per second while ensuring high availability and fault tolerance. Which combination of AWS services and strategies is the MOST appropriate to achieve this goal?
- A
Use Amazon SageMaker Hosting Services with Auto Scaling enabled
- B
Deploy the model on a single EC2 instance with a large instance type
- C
Use Amazon API Gateway to expose the endpoint and integrate it with an AWS Lambda function that invokes your model
- D
Configure an Amazon SageMaker endpoint in multiple Availability Zones
- E
Enable Amazon CloudWatch Alarms to monitor latency and automatically trigger scaling
Show answer and explanation
Correct answers: A, D, E
Explanation
To build a scalable, resilient, and fault-tolerant ML solution for high-throughput real-time predictions, using Amazon SageMaker Hosting Services with Auto Scaling ensures the system can handle varying traffic loads. Deploying the endpoint in multiple Availability Zones ensures fault tolerance and high availability. Additionally, monitoring latency and setting up CloudWatch Alarms allows the system to dynamically adjust resources to meet performance requirements. These strategies collectively address the key requirements for performance, availability, scalability, resiliency, and fault tolerance.
- A. Correct.
Using Amazon SageMaker Hosting Services with Auto Scaling ensures that the model can handle a variable number of requests per second by automatically scaling the instances up or down based on demand, which is essential for scalability and performance.
- B. Incorrect.
Deploying the model on a single EC2 instance is not fault-tolerant or scalable. If the instance fails, the service will be unavailable. It also cannot handle a large number of requests effectively as demand grows.
- C. Incorrect.
Using Amazon API Gateway with AWS Lambda is suitable for lightweight inference tasks but is not ideal for high-throughput, low-latency requirements in real-time predictions. It may introduce additional latency compared to SageMaker endpoints.
- D. Correct.
Configuring an Amazon SageMaker endpoint in multiple Availability Zones ensures high availability and fault tolerance by distributing the workload across zones and avoiding a single point of failure.
- E. Correct.
Enabling Amazon CloudWatch Alarms to monitor latency and trigger scaling ensures that the system can dynamically respond to performance issues, maintaining availability and performance under varying loads.