Dependable Messaging in Sensor Networks

The sensing, computing, communication, and actuation capabilities of wireless sensor networks are increasingly enabling us to observe and control the physical world. Along with opportunities, sensornets bring challenges to the design of systems dependability. In this talk, I will focus on the challenges in the context of messaging.

In sensornets, wireless links tend to be dynamic, both temporally and spatially. Link dynamics are also subject to the impact of application properties such as traffic patterns, which can themselves be dynamic. Thus, a key challenge is how to provide dependable messaging in the presence of link dynamics and varying application properties.

To deal with link dynamics and the impact of application traffic patterns, I propose estimating link properties using application data itself rather than periodic control packets. I will discuss the feasibility of data-driven link estimation, its implications to routing protocol design, and its benefits in terms of energy efficiency and network throughput. I will also discuss how the unique application properties of sensornets enable new transport design for delivering bulk bursty data reliably and in real-time. In addition, I will discuss the architectural issues in incorporating application properties in sensornet messaging.

Besides messaging, I will briefly discuss the challenge of scalable dependability in large scale networks including not only sensornets but also the Internet. I will conclude with some open problems in building dependable and predictable sensornets and networked embedded systems in general.