Some embodiments include a back-end routing engine. The engine can receive traffic data characterizes amount of service requests from front-end servers to a server group of one or more back-end servers that corresponds to a geographical tier in a server group hierarchy. The engine can receive metric measurements in a performance metric dimension for the server group and a performance threshold corresponding to the performance metric dimension and the geographical tier. The engine can estimate a linear derivative between variable traffic data and variable performance metric in the performance metric dimension based on collected sample points respectively representing the traffic data and the metric measurement. The engine can then compute, based on the linear derivative and the performance threshold, a threshold traffic capacity of the server group. The engine can then generate a routing table based on the threshold traffic capacity.