Design Methodologies and Tools for Energy-aware IoT-based Applications

Abstract

Green, sustainable and energy efficient computing terms are gaining more and more attention during the last years. As the number of Internet of Things (IoT) computing devices keeps increasing, energy efficiency is becoming an important requirement, imposing new challenges to software developers. Existing works vary significantly, depending on the abstraction level in which the energy efficiency is treated. On the one hand, from software engineering perspective, there are tools that suggest best practices and guidelines based on empirical studies. On the other hand, embedded system practitioners reduce energy either at hardware level or by making custom transformations at source-code level, using custom techniques, DSPs or memory management optimizations. As applications evolve, there is an increasing need to address energy efficiency at application source code level, beyond general guidelines. Therefore, software tools capable of providing energy consumption estimations and identifying optimization opportunities are vital for assisting developers during the phases of application development. The goal of this dissertation is to introduce the design of application analysis tools that target energy efficiency at the software design level. The introduced tools, coupled with implementation details, are capable of estimating the expected energy consumption of applications running on multiple devices. The proposed tools suggest a number of optimizations to the user with special emphasis on estimating potential gains by acceleration. The presented methodology provides several features, including the combination of static analysis and dynamic instrumentation approaches in order to exploit the advantages of both. The potential use of the proposed methods towards building a tool that focuses on saving energy by suggesting efficient function placements on Edge devices is demonstrated. Finally, a special study of the impact of the suggested optimizations on software development, such as the programming effort, is introduced. The recent increase in demand for IoT embedded systems, such as the control of Heating Ventilation and Air-Conditioning (HVAC) in buildings, motivated our study of a special use-case. HVAC control systems exhibit increased complexity and their operation relies less on human decision-making and more on computational intelligence. The efficiency of these systems is usually limited by the orchestrators’ flexibility to optimize simultaneously multiple, and usually contrary, parameters. Throughout this thesis, we aim to introduce novel solutions for designing model-free orchestrators. Experimental results highlight the superiority of our solutions, as we achieve comparable performance to state-of-the-art relevant controllers without the need of any prior detailed modeling and requiring lower computational and storage resourced without sacrificing the quality of derived results.