Clustering-based solutions for energy efficiency, adaptability and resilience in IoT networks

• Main Author: Shivaraman, Nitin
• Other Authors: Arvind Easwaran
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering::Computer science and engineering::Computer systems organization::Computer-communication networks
• Online Access: https://hdl.handle.net/10356/172951
• Institution: Nanyang Technological University

The Internet-of-Things (IoT) paradigm comprises devices ranging from simple sensors to servers that collect, process, and communicate information between each other. Among them, resource-constrained devices (constrained in processor, memory and battery) are deployed in most IoT applications due to their low cost and high availability. Communication between these devices is the highest energy consuming operation and hence, needs to be efficient to sustain their operations in a network over long periods. Additionally, devices may join, leave or move across the network in an ad-hoc manner. IoT devices must be adaptable to these dynamic changes and maintain a stable load on the network. Owing to their low- cost nature and susceptibility to damage, tampering or battery exhaustion, some devices may become faulty leading to unexpected behavior. IoT networks need to remain functional and resilient even with the presence of such faulty devices. Hence, this thesis identifies three major requirements for IoT networks: energy efficiency, adaptability, and fault resilience. It is very challenging to collectively address all three requirements as they have an orthogonal function to each other. E.g. fault-resilient solutions use complex computations for verification of devices and data that inadvertently reduce the energy efficiency in a resource-constrained environment. Existing solutions achieve one or at most two of the requirements collectively, with no solution incorporating all three requirements. Given the large-scale deployment of IoT devices and their price-sensitive nature, it is imperative to design algorithmic solutions to meet the three requirements as this thesis proposes. In this regard, we propose several communication algorithms to cover the three requirements in this thesis. In the first part of the thesis, we introduce a novel communication algorithm DeCoRIC, that dynamically groups asynchronous devices into clusters and elects a representative among them. DeCoRIC strikes the right balance in the communication load to reduce energy consumption while maintaining connectivity across different devices in the network. Further, unlike most existing clustering solutions which are susceptible to faults and network partitions, we also achieve adaptability and resilience by utilizing message gossips to inform device status among neighboring devices, allowing for alternate connecting paths in the event of a faulty device. We evaluate DeCoRIC on the Contiki simulation platform and compare it against state-of-the-art clustering solutions. We find that DeCoRIC improves the power efficiency of the network by at least 70% and extends its life- time by at least 42% over existing solutions while making the network adaptable and resilient to faults. Further, in the second part, we showcase the practical applicability of DeCoRIC with a use-case of a clustered network of energy resources and propose a charge scheduling algorithm to improve their utilization. In particular, we investigate resource allocation and scheduling of devices in an electric grid network to maximize energy utilization. Through experiments, we found that the mobility of devices allows them to be adaptable by moving to different clusters (geographical location) for energy consumption if the current cluster is overloaded. Hence, we devise an online load scheduling algorithm that exploits different charging modes and mobility of devices to maximize the devices that meet their energy demands. Our solution creates a feasible schedule for the devices in a fraction of time (in the order of minutes) compared to conventional optimization solvers (in the order of hours). We also tested our solution on a real-world electric vehicle dataset and found a 57% improvement in the achieved utility compared to other scheduling solutions such as earliest deadline first or highest-energy demand scheduling. To further improve the energy efficiency and tolerance of the IoT networks, the last part of the thesis builds on DeCoRIC to synchronize the devices and create a timed communication. While the literature on time synchronization addresses energy efficiency and adaptability partially, none of the solutions provide resilience to faults. Therefore, we design a novel time synchronization solution C-sync, that is fault-resilient and adaptable while maintaining microsecond (μs) accuracy. In addition, C-sync uses a byzantine consensus mechanism to identify anomalies in the device communication and ensures correct information is propagated across the network. Further, we introduce a concept of local centers to limit the maxi- mum number of hops a device is located from its time source making the network adaptable. C-sync is tested on a hardware testbed and shown to consume at least 51% lower power than existing state-of-the-art resilient synchronization solutions. The resilience of C-sync to byzantine faults is also demonstrated by a quick recovery of correct time information when faulty information was introduced. The algorithms presented in this thesis collectively achieve an energy-efficient, adaptable and fault-resilient IoT network organically to solve the research gaps despite the orthogonal requirements. This thesis provides the first step in advancing IoT application domains limited by these missing requirements. Applications with a large number of IoT devices such as smart factories and smart cities benefit from the presented adaptable clustered network solution with its real-time monitoring and control. Resources are better utilized and congestion is reduced by the underlying consensus mechanisms while allowing cost-effective equipment maintenance using the fault-resilience capabilities of our solution. Implementations on real-world data and testbeds used for experiments elucidate the practical viability of the solutions for IoT networks.