My research centers around adaptive network resource allocation in wireless/mobile networks and the DiffServ Internet. After observing the limited success of narrowly focused QoS research in the past, I take a system research approach to address the real challenges of realizing network adaptation, which include modelling application adaptation needs, designing foundational algorithms, and applying solutions at access, edge and core networks to tailor different time scale and state management requirements.

Bandwidth utility function is a good model of application adaptation needs. Realizing utility-based network adaptation, however, requires solutions to utility generation, and aggregation. We investigate the formulation of utility function for aggregates of applications [IWQoS'00], and for MPEG video [NOSSDAV'99]. In [PV'99], we devise a utility prediction algorithm alleviating the time scale mismatch between utility generation and network adaptation. On algorithmic design side, we propose a suite of foundational utility-based link sharing algorithms. We introduce utility-fair [IWQoS'98] and proportional utility-fair [IWQoS'00] allocation algorithms, fast calculation of welfare maximization, and a combination of these algorithms in hierarchy [IWQoS'00].

Applying utility-based algorithms at access, edge and core networks requires careful consideration of algorithm time scales and state management. Across access networks where per-device utility states are manageable, we present the utility-based max-min fair algorithm and evaluate its performance in a wireless access network [Mobicom'98]. In addition, we enrich the network support for complex behavior of application adaptations over different adaptation time scales and bandwidth granularities [WINET'01]. In core networks supporting service differentiation, utility-based allocations are kept at the edge, for the purpose of ingress rate regulation [IWQoS'00] and egress peering [JSAC'00]. Within the core network, dynamic provisioning algorithm are designed to to tailor the special need of regulating DiffServ traffic aggregates when their traffic distribution exhibits tree topology [IWQoS'01].

Last, an important issue of application adaptation is incentive engineering, which aims at exerting cooperation from applications and reducing transaction cost. We apply the results from mechanism design in economics theory to the case of wireless/mobile network access. Our design keeps messaging overhead minimum and induces application to truthfully use bandwidth and select service class as needed. Our work are evaluated over a programmable mobile network [mobiware], and a dual purpose emulation/implementation platform for mobile networking [CUBAnet]. These system work allow us to design efficient protocols with minimum signalling requirement on applications.