MLA-C01 Question 242
Single answerYou are a Machine Learning Engineer tasked with deploying an image classification model for a healthcare application. The model must handle real-time inference for 10,000 requests per second with low latency, comply with strict healthcare data privacy regulations (such as HIPAA), and support scalability during peak hours. The company already uses Amazon ECS for containerized workloads. Based on these requirements, which infrastructure would be the most appropriate for deploying the model?
- A
Deploy the model on Amazon SageMaker Hosting Services using an endpoint with auto-scaling enabled.
- B
Deploy the model on an Amazon EC2 instance with custom security configurations.
- C
Deploy the model as a containerized application on Amazon ECS with AWS Fargate.
- D
Deploy the model on AWS Lambda using SageMaker's pre-built container for inference.
Show answer and explanation
Correct answer: C
Explanation
The company's existing architecture uses Amazon ECS for containerized workloads, and the model must handle 10,000 requests per second with low latency while complying with HIPAA. Amazon ECS with AWS Fargate is the most appropriate choice as it supports containerized applications, scales efficiently during peak hours, integrates well with existing infrastructure, and can be configured for strict data privacy compliance. Other options either require significant manual effort, do not scale effectively for the workload, or are unsuitable for the real-time inference requirements.
- A. Incorrect.
Amazon SageMaker Hosting Services is suitable for managed model deployment, but it may not integrate well with existing Amazon ECS infrastructure, and compliance with HIPAA would require additional configurations.
- B. Incorrect.
Amazon EC2 provides flexibility but requires significant manual effort for scaling, managing security, and ensuring low latency, which may not meet the requirements effectively.
- C. Correct.
Amazon ECS with AWS Fargate is an ideal solution because it integrates seamlessly with the company's existing ECS infrastructure, supports containerized model deployment, ensures scalability, and allows compliance with strict regulations like HIPAA through proper network and IAM configurations.
- D. Incorrect.
AWS Lambda is not suitable for real-time inference at the required scale (10,000 requests per second) due to its limitations in execution duration and concurrency, making it an inappropriate choice for this scenario.