MLS-C01 Question 283
Select 2You are working as an ML engineer at a company that hosts a real-time fraud detection application on AWS. The application uses an Amazon SageMaker endpoint for inference and an Amazon RDS database for logging transactions. Over time, you notice that the application’s costs are increasing significantly without a corresponding increase in traffic. Upon investigation, you observe that the SageMaker endpoint is frequently underutilized, and the RDS instance is configured with high Provisioned IOPS but experiences low read/write throughput. What actions should you take to rightsize these resources while maintaining application performance?
- A
Switch the Amazon SageMaker endpoint instance type to a smaller instance class based on the observed utilization.
- B
Enable Amazon SageMaker Multi-Model Endpoint to host multiple models on the same instance to reduce costs.
- C
Reduce the Provisioned IOPS for the RDS instance to match the observed workload.
- D
Switch the RDS database to a General Purpose (gp3) storage type to replace Provisioned IOPS.
- E
Enable SageMaker Elastic Inference to dynamically adjust inference compute resources based on workload.
Show answer and explanation
Correct answers: A, C
Explanation
To rightsize resources, the focus should be on aligning resource configuration with actual workload requirements. In this scenario, the SageMaker endpoint is underutilized, so switching to a smaller instance type reduces costs while maintaining performance. Similarly, the RDS instance with high Provisioned IOPS can be reconfigured to match its low throughput workload, avoiding over-provisioning costs. Other options, such as Multi-Model Endpoints or Elastic Inference, are not applicable to the described situation.
- A. Correct.
Correct. Choosing a smaller SageMaker endpoint instance type based on actual utilization is a cost-efficient way to optimize resource usage while maintaining performance.
- B. Incorrect.
Incorrect. Multi-Model Endpoints are designed for hosting multiple models on the same endpoint, which is not relevant to the scenario since there is no indication of multiple models being deployed.
- C. Correct.
Correct. Reducing Provisioned IOPS to match actual workload reduces unnecessary costs while still providing sufficient performance for the low throughput observed.
- D. Incorrect.
Incorrect. Switching to General Purpose (gp3) storage might save costs but could negatively impact performance if the workload requires consistent high-speed IOPS.
- E. Incorrect.
Incorrect. Elastic Inference is not applicable here as it is intended for attaching GPU acceleration to inference workloads, and there is no mention of GPU usage in this scenario.