Computers were invented to automate the labour-intensive computing process. The advancement of semiconductor technology has reduced the form-factor and cost of computers, and increased their usability. This has gradually introduced computers in various control and automation systems. The further rise of miniatuarized computing devices paves the way for autonomous monitoring using embedded devices. In the last two decades, we observed a huge surge of such monitoring and control systems. These systems are generally termed as the wireless sensor and actuator networks (WSAN). In a WSAN, a number of sensor nodes monitors a deployment area where data collection by humans is either difficult or costly. These devices collaboratively report their sensor readings to a centralized node called the sink. The sink is connected with the Internet, thus delivers the data to the outside world. This way the deployment region can be monitored remotely. Similarly, some actuators can also be controlled remotely through the sink.
In the last decade, the concept of the Internet of Things (IoT) has evolved where any device can be reached by any other device/system/human being from anywhere and anytime. Thus, WSANs can be seen as a precursor of IoT. However, the vision of IoT is not limited to mere remote connectivity. Unlike traditional WSAN, where devices are deployed in remote/critical locations for specific purposes, IoT devices would be integrated into our daily surroundings assisting us in every aspect of life. As the embedded devices are resource constrained, energy and computational efficiency is a major challenge for both WSAN and IoT devices. However, the problem escalates as the IoT devices are expected to perform a number of tasks as opposed to a specific task as performed by classical WSANs. Moreover, the goal of IoT is to take humans out of the control loop or reduce the human intervention as much as possible. This requires devices to exchange data and cooperate among themselves. Thus, IoT devices need to act smartly fulfilling various requirements within its resource constraints.
Every existing and upcoming device and network would be part of the IoT ecosystem. As the number of devices is expected to grow multifold, managing these devices will be a challenge. Especially since these devices are under the control of various entities/organizations. Not to mention that the manufacturers of various devices and their specifications would also vary significantly. To accomplish the vision of IoT these devices need to be able to cooperate and collaborate among themselves even if they are managed differently. This thesis brings forward the concept of virtualization in IoT to tackle the challenges of a global IoT ecosystem.
The first challenge that we tackle is how to virtualize the IoT. We propose a reference architectural model for IoT called DIAT. The reference architecture follows a layered design principle where each layer groups a number of similar functionalities together. This enables easy development of existing and new functionalities of each layer independently. To validate the feasibility and usability of such an architectural model, we developed a system based on a practical IoT-application scenario. To this extent, we developed a controller (iLTC) that operates the heating and lighting systems in an office environment such that these devices operate energy efficiently. At the same time the system ensures a comfortable surroundings for the occupants while eliminating any direct involvement from the occupants.
As WSANs are an integral part of the IoT ecosystem, next, we revisited some of the classic problems of WSANs in the wake of virtualizing the IoT. As energy efficiency is one of the biggest issues in WSANs, we propose a solution to reduce the overall traffic in a network without affecting the quality of data/monitoring. We achieved this by virtualizing the WSAN, which leads to higher cooperation among the devices and a higher operational optimization. We developed the virtual sensing framework (VSF) that exploits the inherent correlation among the sensor nodes to predict sensor readings (virtual sensing). The basic idea is that if a number of nodes are highly correlated, sensor readings from only one of them is sufficient to predict the readings for rest of them. Due to virtualization, such a cooperation among the nodes is possible. This reduces the amount of data transfer within the network, which leads to energy-efficient network operation.
Further, we developed an efficient data collection protocol, called Sleeping Beauty that complements the virtualized sensor network. Based on a centralized schedule, nodes deliver their sensor readings to the sink reliably and efficiently. The accomplishment of a centralized schedule depends on network-wide time synchronization. As the hardware clock of an embedded device drifts significantly within a short time span, we developed a simple self-rectification mechanism such that the overhead of synchronizing the network periodically can be reduced significantly. This technique can be used by any protocol that requires time synchronization other than Sleeping Beauty.
Timely data collection is another desired aspect of IoT as opposed to classical WSANs where latency is generally compromised in order to achieve a higher energy efficiency. We developed a communication mechanism, called Rapid that not only delivers the sensor readings in a fixed time bound, it also reduces the energy consumption. Rapid forms a number of clusters on-the-fly, where the cluster-heads collect data from the cluster-members and send an aggregated packet to the sink. By exploiting the capture effect, Rapid achieves parallelization for intra-cluster communications. Further, it exploits the constructive interference based fast flooding to deliver the aggregated data, which eliminates hop-by-hop flow scheduling. These two factors reduces the overall end-to-end delay of all the flows.
The proposition of this thesis is that by means of virtualization, traditional WSANs can be easily integrated into the grand vision of IoT. We proposed a reference architecture, validated by means of a case study, and developed several amendments to classical WSAN data collection, making it consume less energy and achieve lower latency. We are convinced that virtualization can be applied effectively to other (WSAN) functionality as well. The future of IoT is looking bright.