Mobile ad hoc networks (MANETs) are the core technology that provides the US military with adaptable and reliable battlefield communications. These self-organising networks are ideal for rapidly changing scenarios that require connectivity even under hostile conditions. The excitement and promise that these networks generated transferred to the civilian space as well, fueling over a decade of research. However, they only experienced limited success in this new setting, leading to a fragmentation into several application-oriented sub-fields that dealt with narrower sets of constraints.

In this dissertation, I postulate that the unique properties of these networks makes them much more error-prone than initially considered. Consequently, a focus on improving their capabilities rather than on mitigating their faulty nature made the design of general-purpose MANETs increasingly challenging. I support this argument through an extensive set of experiments that is informed by analysis of the literature, history, and properties of these networks.

Ultimately, my work contributes to the field in three fronts: Firstly, motivated by the multiple challenges that research in this area presents, I designed and built MeshSim, a real-time network simulator. This new platform focuses on code-fidelity and enables me to follow a data-driven experimental cycle. Secondly, I present the Reactive Gossip Routing family of protocols, designed to provide reliable and scalable routing by mitigating the MANET properties that lead to faults. Using MeshSim, I evaluate these protocols experimentally under increasingly harsher conditions and compare their effectivity against the incumbents in the literature. Finally, I demonstrate through experimentation that distance-vector routes are ill suited for MANETs due to a geographical-spreading effect they induce, a result that extends to many routing metrics when used in shortest-path algorithms.