Peer-to-Peer Group Communication for City-Scale Mesh Networks

Abstract

The Internet has become extremely centralized. The benefits of centralization have thus far outweighed the drawbacks, but users today are much more concerned about privacy, and reachability is increasingly threatened by natural disasters, political repression, cyberattacks, and human error. CityMesh provides an answer to this problem, constructing a decentralized mesh network out of wireless access points. To test our unicast routing protocol, we built a discrete-event network simulator using SimPy. However, we make several simplifying assumptions, and unicast is not sufficient for many applications. In this thesis, I show that our simulator nevertheless achieves 67.4% correlation with real data that we collected, and I generalize our simulator for multicast. Specifically, I compose our unicast primitive into multicast trees using three different topologies, and surprisingly find that Steiner trees perform worse than minimum spanning trees on average.