An application reads from a Kinesis data stream. The stream’s shards are configured for normal traffic, but during peak tests the application can’t ingest data fast enough. What is the most cost-effective change to make the stream handle peak traffic?

Increasing the number of shards directly addresses the bottleneck of not being able to ingest data fast enough during peak traffic. Each shard provides a fixed unit of capacity for both reads and writes. By increasing shards, you scale up the stream's throughput, allowing the application to consume data at a higher rate. This is generally the most cost-effective solution for scaling read capacity in a provisioned-capacity Kinesis stream. Installing the Kinesis Producer Library (KPL) helps with efficient data production but doesn't solve the consumption bottleneck. Switching to on-demand capacity mode would handle scaling automatically but is often more expensive than manually scaling provisioned capacity for predictable peak loads and doesn't directly address the read issue without also increasing underlying capacity. Reducing the retention period frees up storage but does not increase read or write throughput.