Question

A financial services company runs a customer-facing chatbot on Amazon Bedrock using Provisioned Throughput with a 6-month commitment for a specific model. During normal business hours, 2 model units comfortably handle steady traffic. However, at the start of each trading day (9:30 AM), a 15-minute burst pushes request volume to roughly 3x normal, and users experience throttling exceptions during this window. The team wants to eliminate the throttling during the burst while minimizing ongoing cost, and they cannot change the model. Which approach best meets these requirements?

Accepted Answer

Keep the committed 2 model units for baseline traffic and configure an application inference profile that routes overflow burst requests to on-demand (cross-region) inference during the peak window.

Answer

Purchase additional committed Provisioned Throughput model units sized for the 9:30 AM peak and keep them running 24/7.

Answer

Switch the entire workload from Provisioned Throughput to on-demand inference to avoid managing model units altogether.

Answer

Replace the committed Provisioned Throughput with batch inference jobs triggered at 9:30 AM to absorb the burst.