A company runs a 3-year project with an application that uses 20 Amazon EC2 On-Demand instances registered to a Network Load Balancer target group across two Availability Zones. The application is stateless and runs 24/7. Normal CPU utilization is about 10%, but CPU rises to 100% during brief busy periods lasting a few hours. Users report slow responses. Which new architecture will resolve the slow-response problem MOST cost-effectively?

The correct option leverages an Auto Scaling group (ASG) to dynamically adjust capacity based on demand, resolving slow responses during peak times. Setting a minimum capacity of 4 instances and purchasing Reserved Instances for them provides a cost-effective baseline for the 24/7 application, as the application's normal CPU utilization is low. The maximum capacity of 28 allows the ASG to scale out to handle the brief, high-utilization periods. The other options are less cost-effective or suitable. Setting minimum capacity to 20 and purchasing RIs for all 20 is expensive given the low normal utilization. Spot Fleets are generally not ideal for 24/7 applications with consistent baseline needs, especially with a maintain request type and DefaultTargetCapacityType to Spot, which could lead to interruptions.