As we cross half a million users, we definitely need to implement auto scaling for the automatic resizing of compute clusters. At this stage, we need to know our requirements in terms of defining min/max pool sizes, the CloudWatch metrics to drive scaling, and using on demand and Spot instances in our auto scaling groups. Over provisioning resources for peak periods will get expensive at these volumes. You can create an auto scaling group for web servers in the web tier across the AZs (if you have been following the suggestions in the sections, the rest of the architectural components are managed services, and scaling them is largely Amazon's responsibility).
At the most basic level, we need ...
