Characterizing Middleware Mechanisms for Future Sensor Networks

Abstract

Due to their unique blend of distributed systems and networking issues, wireless sensor networks (SN) have become an active research area. Most current SN use an arrangement of nodes with limited capabilities. Given SN device technology trends, we believe future SN nodes will have the computational capability of today's handhelds, and communication capabilities well beyond today's "motes", satisfying application demand for greater capabilities for performing computations in-network on higher bit-rate streaming data. We focus on stream-based future SN applications, such as automated surveillance, that perform in-network streaming data "fusion" operations, such as face detection, in a hierarchical fashion to produce high-level inferences to guide actuation decisions, forming a "control loop". Energy will continue to be a primary limiting factor for future SN, so performing in-network fusion in an energy-conscious manner is key to application longevity. There exists a need to study tradeoffs in terms of how much productivity an application can achieve during its lifetime, how application latency and throughput requirements affect both lifetime and productivity, and how various available middleware and device capabilities for performing low-power communication and processing impact these performance metrics.

We evaluate and extend a set of mechanisms used by our recent novel middleware, "DFuse", for application-directed energy management of future SN fusion applications. Our simulation-based evaluation enables modeling a variety of applications, network scales, network layers, and device capabilities to determine how each middleware mechanism impacts performance for a SN context. We extend the set of existing mechanisms (dynamic fusion point migration and optimistic data prefetching) to include local CPU scaling and predictive prefetching to better adapt to bursty workloads while employing an emerging device power management capability. Given these results, we hope to be able to generate a novel model for how to construct a SN in terms of hardware, MAC, routing layer, and tuning parameters for DFuse middleware, given application characteristics and performance requirements.