CloudGuild · Blog · Cheat sheets · Lessons · Certifications
AWS Certified Solutions Architect - Associate Exam Question Walkthrough: Machine Learning Cost Management
Explore this AWS SAA-C03 exam question on optimizing costs for machine learning. Understand the right service choices.
When preparing for the AWS Certified Solutions Architect - Associate exam, candidates often stumble on questions involving service combinations for specific use cases. This question addresses both performance and cost efficiency, crucial for large-scale machine learning models.
The question
A company is using AWS to run a large-scale machine learning model that requires significant compute resources for inference. They want to minimize costs while maintaining performance. Which AWS services combination would provide the best solution?
- A. Amazon EC2 with reserved instances for steady workloads.
- B. AWS Lambda with provisioned concurrency.
- C. Amazon SageMaker with endpoint auto-scaling.
- D. Amazon ECS with on-demand instances.
Think before you scroll
Evaluate each option based on its suitability for handling machine learning inference workloads efficiently while also considering cost reduction. Focus on the capabilities specific to machine learning and the management overhead each service incurs.
The answer
The correct option is C. Amazon SageMaker with endpoint auto-scaling. SageMaker is purpose-built for machine learning. It offers a fully managed service that can scale endpoints automatically to handle varying inference loads, making it cost-effective and efficient.
Why the other options lose
- A. Amazon EC2 with reserved instances for steady workloads. While reserved instances can reduce costs for steady workloads, they lack the auto-scaling feature necessary for dynamic inference demands. This approach may lead to over-provisioning or under-utilization.
- B. AWS Lambda with provisioned concurrency. AWS Lambda is excellent for serverless applications but not ideal for heavy machine learning inference. It has resource limits that can hinder performance for large models, making it a poor choice for this scenario.
- D. Amazon ECS with on-demand instances. ECS is suitable for containerized applications but does not provide the specialized capabilities for machine learning inference that SageMaker offers. Using on-demand instances can also lead to higher costs without the efficiency of auto-scaling.
The concept behind it
The underlying principle here is choosing services that align with specific workload requirements. For machine learning, using services designed specifically for it, like SageMaker, can optimize performance and cost. Understanding how different AWS services fit into various architectural patterns is essential.
Exam trap to remember
Remember: Always choose the service designed specifically for the workload. This often leads to the best balance of performance and cost efficiency.