Over the last few years, two fundamental changes have happened in the domain of Wireless Sensor Networks (WSN). One, the applications have ceased to be edge-alone and have morphed into a edge-enterprise co-design. Two, sensor networks are beginning to be shared by a number of stake holders, with varied requirements. This translates into three key requirements for today's WSNs. One, the networks to be designed should be customizable and repurposeable on short notice. Two, the networks should be sliceable--the ability to deploy and run multiple applications simultaneously. Three, the networks need to be designed in such a way as to enable interaction, collaboration and federation with other networks or agents outside of the network to achieve some common goal. This has led to the re-examination of the application development model for WSNs.
This dissertation proposes an architecture called the fabric model for designing WSN as generic sensing networks, on which applications can be tailored. The model focuses on standardizing services and APIs that are common to most sensor fabrics, so that the common services can be reused across fabrics.
We illustrate the usefulness of the fabric model by specifying and building three different fabrics, but by reusing a number of common services. The three fabrics, Kansei--a sensor testbed with a wired control channel, PeopleNet--a mobile wireless network and VillageNet--an intermittently-connected mobile network, use the same fault-tolerant fabric manager and I/O services. As one of the key contributions we present the architecture for a fault-tolerant, scalable and autonomous fabric manager that can manage thousands of sensor nodes.
While we reuse a number of services in PeopleNet and VillageNet from Kansei, a few significant challenges remain in designing mobile WSN fabrics. An energy constrained multi-hop mobile network, such as, PeopleNet and VillageNet, needs an efficient and reliable routing protocol. Thus, we make two key contributions in the mobile routing space. We have designed & implemented the Asymmetric Event-driven Routing (AER) service, which provides energy efficient messaging service in a slowly mobile network, and the Reliable Energy Aware Predictive Routing (REAPER) service, which provides end-to-end messaging for intermittently-connected mobile networks.
The fabric model provides a number of advantages in designing customizable fabrics and its services based design lends itself to WSN federations naturally. But, a number of other challenges remain in federating WSN fabrics. We present KanseiGenie, a GENI-compliant software architecture for federating geographically separated sensor fabrics and to provide the user with a common interface to program across the federated fabrics. KansiGenie, builds on top of the fabric model and tackles issues related to resource and experiment specification in federating sensor fabrics. We also demonstrate through WEAVE--a domain specific service--how a federated application can be stitched using multiple sensor fabrics.
As sensor networks become ubiquitous and federated, we hope and believe that our work in this dissertation will serve as one of the cornerstones, which will encourage further research in the design and integration of sensor networks.