The second approach to unknown or unpredictable patterns is on-demand mode, where we do not control the capacity but instead rely on AWS to scale the service on demand to meet the traffic volume automatically. This can scale to thousands and even tens or hundreds of thousands of reads and writes per second completely seamlessly. Essentially, we point our application toward a DynamoDB table and then the service determines the sustained performance of our application and allows us to peak at double the sustained rate.

For example, if your application is writing at a sustained rate of 10,000 capacity units, the service will allow the handling of up to 20,000 capacity units of peak traffic. Once our application's sustained rate increases or decreases, the peak rate will be automatically adjusted up or down accordingly.