Predictable and Dependable Low-power Wireless Networks

Опубликовано 22 июня 2016, 0:55

By integrating components for sensing, communicating, computing, and actuating, Cyber-Physical Systems (CPSs) enable applications to monitor and control physical processes. The use of wireless, battery-powered devices unlocks tangible benefits in these emerging CPS settings. However, wireless communications is notoriously unreliable, and small form factors and battery-powered operation impose constraints on the computation and communication capabilities. These issues present severe threats to CPSs, whose safety-critical nature calls for two key features from the networking substrate: (1) predictability at design time, to analyze the behavior of the system before it is deployed, and (2) dependability at run-time, to ensure the correct functioning of the system despite failures. So, in this talk, I will pose the following question: Is it possible to provide predictable and dependable end-to-end communication guarantees on top of unreliable multi-hop low-power wireless networks? I will explain why it is impractical to address this question using conventional low-power wireless stacks, and describe our work on designing and building a clean-slate communication architecture from the ground up to overcome this problem. In particular, the proposed architecture provides real-time and virtual-synchrony guarantees on the end-to-end delivery of messages - features that were previously deemed impossible to achieve in these networks. Real-time communication is crucial for the correctness of many CPSs, and virtual synchrony enables the design of fault-tolerant CPSs using state machine replication. Moreover, the architecture allows for predicting the network’s end-to-end performance using simple models with unprecedented accuracy, which can greatly aid in the design and verification of CPSs.