A memory-heavy application runs on EC2 instances in an Auto Scaling group behind an Elastic Load Balancer. The admin needs autoscaling to react based on how many users are connected to the app. Which solution will meet this requirement?

The correct solution is to create a scaling policy that uses the ELB metric ActiveConnectionCount from Amazon CloudWatch. This metric directly reflects the number of client-side connections that are established to the load balancer, which closely correlates with the number of users connected to the application. Auto Scaling can then use this metric to dynamically adjust the number of EC2 instances, ensuring the application can handle varying user loads. Using the memused metric from CloudWatch is incorrect because memused is an instance-level metric, not an ELB metric, and it measures memory utilization on the EC2 instances, not the number of connected users. A scheduled scaling policy is not ideal because user traffic is often unpredictable, making a reactive policy more efficient than a time-based one. Installing a script on the ELB to export custom metrics is not possible; ELBs are managed services and do not allow custom script installation.