Abstract

The Internet of Things is a new paradigm that allows an enormous number of devices i.e., sensors, actuators, RFID tags, etc. to cooperate to reach a common goal. The wireless sensor networks as the key components of the IoT are extensively being used in various domains and applications. However, in traditional WSNs, applications are embedded into the sensor, precluding them from being re-used by other applications. Therefore, the sensor become application-specific and task-oriented devices with increased deployment and maintenance costs. To cope with these issues, a viable approach is to apply virtualization to WSNs. Virtualization abstracts the physical sensing capabilities of the sensors into logical units, allowing them to be reused by multiple applications. WSN virtualization can be done at either node- and/or network-level. However, virtualization challenges the energy consumption in WSNs. Thus, inefficient application provisioning can have a drastic impact on the energy consumption of the WSNs, leading to faster depletion of sensor nodes’ batteries. This thesis proposes algorithmic approaches to tackle the key challenges related to energy consumption in virtualized IoT-based networks.

The first challenge faced by virtualized IoT networks is the energy efficiency in dynamic task assignment considering node-level virtualization. Addressing this challenge is critical for energy efficiency. Considering that the IoT devices (e.g., sensors) need to interact and exchange messages, the second challenge is the problem of energy efficiency in dynamic network embedding in virtualized IoT networks considering both node- and network-level virtualization. Yet, another challenge is energy-efficient distributed task assignment in virtualized IoT networks. The IoT nodes are constrained devices with limited available energy and processing capabilities. Thus, it is not always feasible to have powerful nodes in the network to execute the algorithms.

To tackle the first challenge, we modeled the problem using integer linear programming and proposed a heuristic to solve the problem. For the second challenge, after modeling the problem using ILP, we proposed our Dynamic Network Embedding heuristic to solve the problem in an energy efficient manner. When it comes to the third challenge, we modeled the problem using non-cooperative game theory and proposed an energy-efficient heuristic to solve the problem.