Market-Based Resource Management of Edge Computing Systems and IoT Architectures

Abstract

The rapid growth of IoT has exploded the number of devices connected to the Internet, a number which keeps increasing in high pace. This has led to an enormous amount of collected data and information, traditionally offloaded to cloud computing infrastructure.

However, cloud servers and datacenters are geographically centralized, situated far from the end devices and as a consequence, real-time and latency-sensitive computation services often endure large round-trip delay, network congestion and service quality degradation. %For instance, healthcare companies are reluctant to rely on on third-party cloud servers to store or process their sensitive medical data. Moreover, in data-sensitive domains such as Healthcare, privacy and confidentiality concerns have been raised owned to the storage of data to third-party infrastructure. These reasons (lower latency and higher privacy) have driven the Edge computing paradigm, where the required computation is pushed to the Edge of the IoT network in order to alleviate the dependency on cloud infrastructure and enhance the privacy of identifiable personal data.

The presented work regards the well-established Edge computing architecture of IoT nodes and Gateways, where a portion of the tasks of the IoT nodes are/can be offloaded to the IoT Gateway. In this setup, resources including available CPU, memory and communication bandwidth on both IoT nodes and Gateways are limited and a portion of them is shared. Thus, an efficient resource management mechanism is required to dynamically allocate the shared Gateway resources and to designate the operating configuration of IoT nodes.

In this diploma thesis, \textit{DMRM} is presented: a fully distributed market-based resource management system on IoT architectures. The basic idea of the algorithm is based on fundamental economic and pricing models. More specifically, Supply and Demand model, Consumer Perceived Value Pricing and Smart Data Pricing are applied in this study.

In the beginning, there exists an introduction on IoT, Cloud, Fog and Edge Computing and the connection between computer science and economic theory. Afterwards, there is related work and similar research presented and basic economic models are analyzed and compared among them. Additionally, there is presented a brute-force, an Oracle prediction and a Simulated Annealing solution to the resource management problem. Next, there is an extended analysis of \textit{DMRM} and its mechanisms and, after that, it is being studied, evaluated and compared with other solutions. The algorithm is applies on Raspberry pi 3 Model B, Intel Galileo 1 and Tegra X1. Finally, the conclusions of this diploma thesis are summarized and ideas for future research are proposed.