Energy and carbon aware scheduling in supercomputing

At an early stage of information and communications technology and high-performance computing, performance and reliability were two important factors in research and development. Energy consumption was not considered as a serious topic, since the technical characteristics of hardware and software were limited and the amount of computing nodes in a computing cluster, i.e., a data centre was small. Gradually the situation has evolved a lot: nowadays there are multiple data centres located in geographically diverse locations and the software has become more complex. Modern data centres are equipped with a large amount of computing nodes having vast computing power. Consequently, energy consumption has become a major topic. This work presents two algorithms for optimizing energy and emissions in high-performance grid computing, in which multiple data centres are interconnected to each other. The algorithms are validated both in simulation and testbed environments. The effect of various parameters to energy and emission savings are studied and the performance of the algorithms is compared to commonly used default algorithms. Our simulation and testbed experiments show that the developed algorithms are able to reduce energy consumption and emissions drastically without significant increase in job turnaround or wait time.