A method for routing service requests to service instances in a service providing infrastructure is disclosed. A service request is received. Computing resource vectots for at least two service hosts are computed, and each of the service hosts have a corresponding service providing infrastructure. A resource model for the service request is retrieved. The retrieved resource model are matched to each of the resource vectors to identify a best-fit resource vector. The service request is routed to a selected service host associated with the identified best-fit resource vector. Processing of received service requests is monitored. Individual resource components in the service hosts which are consumed during the processing are identified. Resource models for the service requests are produced based upon the identified individual resource components in the service hosts. A time series model is computed for each of the service requests based upon the identified individual resource components.