CloudGuild · Blog · Cheat sheets · Lessons · Certifications
AWS Certified Solutions Architect - Associate Exam Question Walkthrough
Explore a common decision-making trap in the AWS SAA-C03 exam regarding cost-optimized machine learning deployments.
An organization wants to deploy a machine learning model in production on AWS while minimizing costs. The model is not expected to receive constant traffic. Which service should they use to achieve this?
A. Amazon SageMaker Endpoint
B. AWS Lambda with Amazon SageMaker
C. Amazon EC2 Spot Instances
D. Amazon SageMaker Batch Transform
Think before you scroll
Consider how each option handles cost and traffic variability. Focus on services that offer flexibility and event-driven architectures to minimize expenses when the model is not in use.
The answer
The correct option is B. AWS Lambda with Amazon SageMaker. This choice allows the model to run only when needed, significantly reducing costs through event-driven execution.
Why the other options lose
A. Amazon SageMaker Endpoint: This service is always on, leading to higher costs. It is not suitable for models with variable traffic since it incurs charges even when idle.
C. Amazon EC2 Spot Instances: While cost-effective for certain workloads, Spot Instances are not ideal for machine learning inference due to their unpredictable availability and potential interruptions, which can disrupt model execution.
D. Amazon SageMaker Batch Transform: Although it is a viable option for batch processing, it is not the best fit when low-latency, real-time inference is needed. It works well for scenarios where immediate responses are not required.
The concept behind it
The key principle here is to choose architectures that align with traffic patterns. For applications with fluctuating demand, event-driven models like AWS Lambda paired with SageMaker can help maintain cost efficiency. Understanding the nature of your workload is crucial in selecting the right service.
Exam trap to remember
Always consider the implications of running services continuously versus event-driven execution to avoid unexpected costs.