A company runs a highly available web application on Amazon EC2 instances behind an Application Load Balancer. CloudWatch metrics show that, as traffic increases, some EC2 instances become overloaded with many outstanding requests. The overloaded instances have both a higher number of requests processed and longer response times than other instances. The company does not want new requests routed to instances that are already overloaded. Which solution meets these requirements?

The correct solution is to use the least outstanding requests algorithm based on the RequestCountPerTarget and ActiveConnectionCount CloudWatch metrics. The Application Load Balancer (ALB) can be configured to route new requests to the target with the fewest outstanding requests. This directly addresses the problem of overloaded instances having many outstanding requests. RequestCountPerTarget and ActiveConnectionCount are relevant metrics that contribute to determining the load on an instance. Round robin routing would distribute requests evenly without considering current load, potentially sending new requests to already overloaded instances. While RequestCount and TargetResponseTime are related to instance performance, they are not directly used by the ALB's built-in least outstanding requests algorithm to determine which instance has the fewest outstanding requests.