A load balancing scheme is described for a server cluster. An application identifier is received by the cluster from a client desiring to connect to the application. Each one of the servers in the cluster independently hashes the application identifier into a unique number. This unique number is used to determine a preferred one of the servers for the connection. The servers then each calculate a priority code using the unique number, a load level for each server, and/or a node identifier assigned to the servers in the cluster. If the preferred cluster server is available, it transmits its priority code to the client before the other servers transmit their priority codes. Otherwise, each of the available cluster servers transmits their priority codes to the client. The client selects the cluster server with the most favorable priority for making the connection.