We prove that under a many-server fluid scaling and overload conditions, a server-scheduling policy that assigns priority to classes according to their index c i μ i /θ i , is asymptotically optimal ...