From TimeSync to EmStar: What's really hard in sensor networks?

Abstract

Recent advances in miniaturization and low-cost, low-power design have led to active research in large-scale networks of small, wireless, low-power sensors and actuators. Sensor networks have enormous potential, but their design is challenging in many unexpected ways.

Our research started out in time synchronization -- an important service in any distributed system, but particularly crucial in sensor networks. We developed and implemented several new approaches that better support the unique requirements of this domain. For example, Reference-Broadcast Synchronization achieves high precision at low energy cost by leveraging the broadcast property inherent to wireless communication. A new multi-hop algorithm allows RBS timescales to be federated across broadcast domains.

When the time came to apply these new methods to academic and commercial systems that needed synchronization, we learned many interesting lessons. Perhaps most surprising: in sensor networks, applications that, on the surface, might seem to hinge on problems like time synchronization actually were hard for completely different reasons.

Based on this experience, we developed EmStar, a software framework for sensor networks. EmStar's goal is to make the hardest problem easier: writing software for an environment that is inherently unpredictable, dynamic, difficult to model, hard to observe, and always subject to Murphy's Law.