Robustness, Stability, and Gains of Utility Maximization Algorithms for Mobile Ad Hoc Networks

Abstract

Cross-layer designs (CLDs) have the potential to better utilize network resources than a traditional layered protocol architecture, providing better communication services to end-users. This is particularly relevant in MANETs, which are characterized by a scarcity of network resources, in particular bandwidth. While many papers have demonstrated the potential performance benefit that CLDs can provide, their evaluation is often done under idealized scenarios, such as instantaneous message transmission times, complete topology knowledge, and/or zero message loss rates. Nodes in real MANETs do typically experience message losses, message delays, and have at best a partial and most likely inaccurate knowledge of the dynamic network topology. Our work focuses on realistic evaluations of a specific CLD approach. We propose a robust and stable version of a network utility maximization algorithm that optimizes the medium access probabilities at the MAC layer jointly with the end-to-end sessions rates at the transport layer. We report on stability tests with different link failure scenarios in case of synchronous and asynchronous message updates, convergence speed, and link utilizations. The results show that the algorithm converges even with a very high rate of link transmission failures and can tolerate asynchronous parameters updates. The CLD fully utilizes the link capacities and provides a network capacity gain of up to 250 %. The proposed modified CLD approach is therefore both beneficial and practical for real MANETs.