Google Professional Machine Learning Engineer Question 322
Select 3Google Cloud PlatformYou are tasked with deploying a machine learning model to serve predictions for a high-traffic e-commerce application. The application experiences unpredictable traffic spikes during promotional events. Which considerations should you prioritize to ensure scalable and reliable online model serving on Google Cloud?
- A
Use Google Kubernetes Engine (GKE) with horizontal pod autoscaling enabled to dynamically adjust resources based on traffic.
- B
Deploy the model on a single Compute Engine instance with preemptible VMs to reduce costs during low traffic.
- C
Integrate Cloud Load Balancing to distribute incoming requests across multiple model serving instances.
- D
Implement prediction request batching to optimize resource utilization during periods of high traffic.
- E
Manually scale the infrastructure before promotional events to handle anticipated traffic spikes.
Show answer and explanation
Correct answers: A, C, D
Explanation
To scale online model serving effectively for high-traffic applications, you need solutions that dynamically adapt to changing traffic patterns. GKE with horizontal pod autoscaling ensures automatic scaling, while Cloud Load Balancing distributes the load to prevent bottlenecks. Additionally, batching requests can further optimize resource utilization during peak usage. Manual scaling and preemptible VMs are not reliable or scalable solutions for unpredictable traffic spikes.
- A. Correct.
Using Google Kubernetes Engine (GKE) with horizontal pod autoscaling allows your deployment to adjust resources dynamically. This ensures that the infrastructure can handle traffic spikes without manual intervention, making it a reliable and scalable solution.
- B. Incorrect.
Deploying the model on a single Compute Engine instance with preemptible VMs is not suitable for high-traffic scenarios, as preemptible VMs can be terminated at any time, leading to potential downtime.
- C. Correct.
Integrating Cloud Load Balancing distributes incoming requests across multiple instances, which enhances scalability and ensures reliable service during traffic spikes.
- D. Correct.
Prediction request batching can improve resource utilization and throughput during periods of high traffic, reducing latency and resource strain.
- E. Incorrect.
Manually scaling the infrastructure is not an efficient or scalable solution, as it requires human intervention and may not respond quickly enough to sudden traffic spikes.