A Colored Petri Net - based model for Internet of Things networks

Abstract

The Internet of Things (IoT) has been a significant driver of transformation in the global economy, with an estimated impact of up to US$ 12.6 trillion by 2030. By 2025, it is projected that more than 5 billion devices will adopt 5G LTE Narrowband Internet of Things (NB-IoT) technology. IoT networks are characterized by their heterogeneity in hardware and software, requiring runtime adaptation with minimal human intervention. However, IoT devices typically operate under constrained power supplies, necessitating efficient energy management strategies, even when energy harvesting devices are employed to extend their lifespan and ensure energy-neutral operation (ENO). Transceivers play a crucial role in IoT devices, known for their significant power demands. Channel impairments and small buffers in transceivers can contribute to increased energy consumption, particularly due to retransmissions. Furthermore, transceivers lacking packet filtering mechanisms and checksum capabilities require additional CPU processing to handle corrupted or irrelevant packets. Additionally, the microcontroller unit (MCU) is another power-intensive component in IoT devices, with multiple power states and input supply pins, that must be considered in the analysis of IoT networks. Achieving a balance between performance and energy consumption presents a significant challenge in IoT networks. Therefore, the development of solutions that enable proper dimensioning, configuration, and analysis of IoT networks prior to deployment is crucial for their success. In this respect, this thesis presents a Colored Petri Net (CPN)-based model for IoT networks. It allows evaluating network nodes individually, considering CPUs from different technologies, with several power states, multiple input supply pins, multiple transceivers, capacitor or battery-oriented devices as power supplies, energy harvesting devices, and non-ideal channels. The model encompasses metrics such as throughput, network lifetime, expected number of depleted nodes, expected number of nodes employing low power mode mechanisms, expected number of nodes that fails while transmitting, expected number of nodes transmitting simultaneously, availability, mean time to failure (MTTF), mean time between failure (MTBF), mean time to repair (MTTR), and energy consumption. Validation results with real measurements show that the proposed model presents relative errors lower than 0.34% and 1.55% for CPU and transceiver energy consumption, respectively, and close to zero for the throughput. These results show the suitability of the proposed model to assist the system designer in the analysis, dimensioning and configuration of IoT networks.