Search

To date, a wide range of industries leverage the Internet of Things (IoT) to improve efficiency and productivity. Briefly, an IoT network consists of wireless sensor devices that are used to monitor an environment, provide coverage, or collect data of targets. The collected data can then be processed before it is uploaded to a sink/gateway or the cloud to be used by various applications. The full potential of IoT, however, is limited by a number of issues. First, most IoT networks are designed for a specific application. This motivates the use of virtualization technologies, where the same network substrate runs multiple applications concurrently. In particular, devices or network operators virtualize resources such as energy, memory, and/or computational time/cycles, which in turn allows these resources to be shared by virtual network functions (VNFs) belonging to different applications. The second issue concerns energy. Specifically, the limited amount of energy on sensor devices governs the amount of data collected or processed by an IoT network. To this end, this thesis considers a number of approaches to overcome the energy limitation of devices. The first approach is to employ imprecise computation to trade-off the quality of computation result with the energy consumed by devices. Another approach is to apply energy harvesting techniques to harness ambient energy, e.g., solar. In this respect, a key consideration is battery property, meaning devices have to control their charging and discharging operations to avoid non-ideal properties, such as memory effects.