A complex user-facing computer application often has run-time dependencies on other computer applications. The other computer applications may, in turn, have run-time dependencies on still other applications. A supporting application might run on multiple hosts and a particular host might be chosen by a higher-level application in order to meet requirements such load balancing or reliability. In order to facilitate intelligent choices by higher-level applications in the system, each server in the system is responsible for generating a performance capability or health score that reflects the health of local components and the health of all servers on which the given server has a direct run-time dependency. A particular server's generated health score is advertised to any other server that has a direct run-time dependency on the particular server. Decisions about which of alternative lower-level servers to use in a servicing a client request are made using a routing or hop-at-a-time approach.