A Survey on Automatic Source Code Transformation for Green Software Generation

Abstract

Hardware rapidly develops, evolving incessantly towards new and more efficient architectures. We are witnessing a significant shift in the history of computing towards parallelism, which raises a new major concern in computing: the energy consumption of computing devices, from small battery-based devices to large data centers forming the cloud. The default electronic computing device today is a parallel machine. To improve the performance, computing devices include more and more processors and cores, as well as larger memories, consequently increasing their energy requirements.

There are billions of computing devices in the world, many of them operating 24 h a day, 7 days a week. This, coupled with the fact that servers increasingly consume more energy, makes communication and information technologies a substantial contributor of emissions in the planet. Software is present in all computing devices and ultimately determines their energetic behavior. Consequently, any modern software requires the minimization of the energy consumption of the device while running.

Unfortunately, writing software is complex and costly, and creating low consumption (or green) software is a challenging task that requires extensive programming experience and deep knowledge on architecture. Balancing between software energy efficiency and functionality might require trade-offs, potentially compromising user experience in certain situations. This chapter surveys the main existing works related to the automatic generation of green software, and emphasizes the main existing challenges in the field.