On utility-based network management

In this paper we discuss the viability of utility-based abstractions for network management. These can be used for estimation and optimization of the network performance. We consider utilities that are determined by the application type, but also take into account changing user and provider context. These metrics incorporate technical performance characteristics, economic and personal considerations of network stakeholders. We discuss on approximations of utility functions and related optimizers. Utility approximations are particularly attractive option to avoid the considerable signaling and computational overhead that otherwise can be imposed by overly detailed specification of optimization problems. As examples we consider simplified versions of the class-based advanced queue management, and the workload/cost balancing between data centers. Our experiments show that in some cases up to 30% performance gains can be achieved compared to the standard approaches.