Primal and dual approaches to distributed cross-layer optimization

Several approaches for cross-layer design, e.g., coordinating the traditionally separated layers in wireless networks, have been proposed. However, protocols that are close to achieving the performance bounds are still lacking. We propose three distributed algorithms for joint congestion control and resource allocation in networks with variable capacities subject to a global resource constraint. Examples include spectrum assignment in wireless networks and wavelength allocation in optical networks. For scalability, we impose the additional constraint that nodes can only negotiate and exchange resources with their neighbors. The proposed algorithms consist of two complementary approaches based on decomposition techniques, in which congestion control and resource allocations are performed on different time-scales. Two of the algorithms can be shown to converge without network delays.