Data simulation has an important role to play in many areas of science, in order to learn new things. But increasingly, sufficient supercomputer capacity is needed to lay the right foundations for scientific advancements. The next step up in computational power is exascale computing. It promises to enable simulations with incredible accuracy, providing detailed insights into many important processes. These supercomputers of the future will be able to carry out at least 10 trillion operations per second. But achieving such computational performance using conventional technology is only possible with extensive electricity consumption. As a result, not only is computer performance becoming a critical factor in supercomputers, so is energy efficiency. This is the starting point for an European project called Exa2Green, which is being implemented by Steinbeis-Europa-Zentrum (SEZ) in collaboration with other scientific partners.
Under the coordination of the computing center at Heidelberg University, a team of mathematicians, IT experts, physicists, and engineers is looking at the challenge of improving the energy efficiency of supercomputers. The aim of the project is to develop energy-saving computation methods and corresponding technologies. The main areas of emphasis in research are: the development of tools for monitoring energy consumption and creating profiles; the development of a new measurement technique for use in the quantitative analysis of the energy profiles of algorithms; the application of energy-aware, elementary system kernels; the development of program libraries for energy-efficient linear algebra; the implementation of an energy-aware simulation plan for supercomputers.
The project is just about to draw to a close after almost three years and the team of researchers has much to show for its efforts. Over the course of the Exa2Green project, a new kind of measurement device was developed which makes it possible to measure the energy consumption of individual components inside computers. Previously it was almost impossible to assess the power consumption of parts like processors, memory, hard drives, or graphic cards. The new measuring device is compact enough to fit inside the housing of a computer yet still delivers accurate results. This will help scientists and technical experts to localize inefficient energy consumption and optimize application codes. The team also developed a model for mapping and predicting the time, power, and energy requirements of several elementary computer kernels, also examining the energy consumption and performance of a weather forecasting model called COSMO-ART on a variety of supercomputers. The knowledge the experts acquired as a team on the development of energy-aware applications was then used to apply corresponding algorithms. The team developed techniques for measuring and minimizing the energy consumption of individual computers as well as major computer centers. It also investigated a platform that measures electricity performance to understand energy use of the COSMO-ART modeling system in an attempt to use a more energy-aware procedure.
Prof. Dr. Vincent Heuveline, coordinator of the Exa2Green project at Heidelberg University, emphasizes how important collaboration has been for the success of the project until now: “Our interdisciplinary project consortium comprises partners from the fields of High Performance Computing, computer science, mathematics, physics, and mechanical engineering. By each contributing their specific skills, these partners are making it possible to tackle key research topics.” The project team is confident that their work is making a valuable contribution to improvements in the machines of today and the realization of energy-efficient exascale computing in the future. “Besides the obvious intrinsic value of new, energy-aware algorithms, which make it possible to solve the same problems only with lower levels of energy consumption, this work is also extremely important when it comes to a new, holistic prospect of creating energy-aware computer systems. Our prime goal is not just to be able to make better use of existing machines but also to develop hardware and algorithms of the next generation and point this in the right direction to spawn the exascale supercomputers of the future,” continues Heuveline.
SEZ assisted Heuveline during the project bidding process and the contract negotiation with the European Commission. As project partner, SEZ is also responsible for the project and knowledge management partner and looks after issues relating to intellectual property rights and training within the consortium. It also helps with the dissemination of project results.