You manage a global external Application Load Balancer serving a stateless web application backed by two managed instance groups (MIGs) in different regions. During a regional traffic spike, one MIG becomes overloaded and returns high latency, even though the second MIG has spare capacity. You want the load balancer to distribute new connections based on backend CPU load and to stop sending new requests to backends that are already near their capacity limit. Which backend service configuration should you apply?