Google Professional Machine Learning Engineer Question 343
Select 3Google Cloud PlatformYou are running a machine learning model on Vertex AI Prediction to serve online predictions. During peak hours, the request throughput increases significantly, causing latency issues. You want to ensure that the serving backend can handle this increased load automatically without manual intervention. Which of the following steps should you take to scale the serving backend based on the throughput?
- A
Enable autoscaling on the deployed model's endpoint configuration.
- B
Set a static number of nodes to handle the peak load.
- C
Configure minimum and maximum node counts for the model's deployment.
- D
Utilize Vertex AI's batch prediction feature to handle the high request volume.
- E
Monitor and adjust autoscaling policies using Cloud Monitoring.
Show answer and explanation
Correct answers: A, C, E
Explanation
To handle dynamic request throughput in Vertex AI Prediction, enabling autoscaling allows the serving backend to automatically adjust resources based on demand. Configuring minimum and maximum nodes ensures appropriate scaling boundaries. Monitoring and adjusting autoscaling policies ensures the solution remains optimized over time. Static node allocation and batch prediction are not suitable for real-time, throughput-sensitive scenarios.
- A. Correct.
Enabling autoscaling allows the serving backend to scale the number of nodes dynamically based on the request throughput. This is a key feature of Vertex AI Prediction for handling variable loads.
- B. Incorrect.
Setting a static number of nodes will not allow the system to dynamically adjust to changing throughput, leading to inefficiencies during both peak and low usage periods.
- C. Correct.
Configuring minimum and maximum node counts ensures that the autoscaler has defined boundaries to scale within, preventing over-provisioning or under-provisioning of resources.
- D. Incorrect.
Batch prediction is designed for asynchronous, large-scale inference jobs rather than real-time, online prediction use cases. It is not applicable to this scenario.
- E. Correct.
Monitoring and adjusting autoscaling policies using Cloud Monitoring helps to optimize and fine-tune the autoscaling behavior over time, ensuring it aligns with the application's performance and cost requirements.