Network Lifetime Analysis of Data Collection Protocols

When a sensor network is deployed, we fundamentally care about three main outcomes: to obtain as much data as possible (high delivery rate), to obtain data as fast as possible (low latency), and to obtain data for as long as possible (long lifetime). This last metric, called network lifetime, is of great importance and has been widely investigated because sensor nodes are usually battery-operated. However, there is a gap between the many theoretical studies and the very few empirical ones. The aim of this thesis is to bridge that gap. To achieve our aim, we analyze two well-known data collection protocols: one based on shortest-path trees, called CTP; and the other based on opportunistic routing, called ORW. Both protocols have advantages and disadvantages with respect to the network lifetime. On the one hand, CTP reduces the total number of transmissions in the network, but uses an expensive communication primitive and does not care about load balancing. On the other hand, ORW has the exact opposite characteristics, good load balancing with an efficient communication primitive at the cost of increasing the total number of transmissions. There is hence an open question to solve: which protocol provides longer lifetimes? We tackle the problem from an analytical and a practical perspective. For the analytical part, we improve the accuracy of current energy models for CTP and develop a new energy model for ORW. Our models for CTP are up to 95% more accurate than the state-of-the-art. For the empirical part, we evaluate both protocols on a public testbed with 100 nodes. Our analytical results show that ORW has longer lifetimes than CTP for high density networks, and that this advantage should vanish in low density networks. Our empirical results validate that ORW is indeed better than CTP under high densities, but for lower densities, our experiments actually show that ORW performs significantly worse than CTP. We show that this unexpected behaviour (according to the model) is due to some inherent flaws in the implementation of ORW.