Service-aware network management is a key issue for the current and next generation service-oriented networks. In this work, we focus on improving the QoS user experience (QoE) of Internet subscribers connecting to heavily loaded servers. Firstly, we advocate an innovative architecture for fine-grained admission-control of Internet servers subjected to multiple flow-based sessions. The proposed architecture is based on original concepts which meet the constraints of multiple-flow based sessions by explicitly identifying the flows pertaining to a single subscriber’s session. Secondly, based on an economically-justified service model, we investigate novel session-aware admission-control for improving providers’ profitability. Our fundamental observation is that it is sometimes desirable to reject some customers’ new sessions so that others may be completed and thereby generate some revenue for the service provider. We evaluate the performance of the advocated model by conducting advanced simulations where we consider realistic CBR voice traffic. Results demonstrate that responsive session-aware admission-control improves service provider’s benefits. For a satisfactory operator’s revenue, session-aware admission-control improves the success rate by more than 69 %, compared to standard request-aware admission-control.