SFB 912: HAEC - Highly Adaptive Energy-Efficient Computing

数据中心和通信基础设施等主要互联网组件的能源消耗和由此产生的生态影响正在达到令人震惊的水平。2012年，信息和通信技术占全球电力消耗的4.6%，年增长率为7%，相当于每十年翻一番。通信网络、计算机和数据中心对上述能源消耗同样负有责任。协作研究中心（CRC）HAEC（Highly Adaptive Energy-Efficient Computing）的愿景是研究在不影响高性能的情况下实现高能效计算系统的技术。当然，提高能效的一个直接方法是降低每个硬件组件的能耗。然而，同样重要的是要了解软件系统如何适应硬件，反之亦然。今天，软件是在没有能量感知的情况下编写的，并且不可知地映射到并行机的通用硬件配置上。由于计算问题需要一定量的特定问题处理和相互通信，因此高度自适应的硬件系统可以根据软件系统的需求优化其配置，从而可以产生更高的效率水平。为此，应用程序和硬件系统的系统状态也需要在运行时进行监控和考虑。在这方面，HAEC将集中研究更大的服务器系统，通过从应用到硬件的创新，以实现高度适应性能源效率计算（HAEC）的综合方法，在涉及的各个技术层面，包括硬件，计算机体系结构和操作系统，软件建模以及应用建模和运行时控制层面，探讨这个问题。一个新的概念进行了探讨，HAEC盒，涉及光和无线芯片到芯片的通信。这允许一个新的水平的运行时自适应性，创建一个平台，灵活地适应硬件配置的计算问题的需要。为了在软件级别利用这种能力，研究了一种分层的能量控制环路基础设施，该基础设施负责根据当前硬件状态以及上下文感知的应用要求来控制能量权衡。因此，港灯循环再用系统是首个尝试，透过由硬件组件至应用软件的各个层面，以综合方法达致高适应性及能源效益。在第一期，我们在新技术方面取得重大进展，并为实现港灯的愿景提出构思，从而为这个愿景奠下基础。在第二阶段，我们采用了这些想法，展示了革命性的方法，大大降低了HAEC的通信功耗，

The energy consumption and the resulting ecological impact of main Internet components as data centers and the communications infrastructure are reaching alarming levels. In 2012, information and communication technology accounts for 4.6% of the worldwide electricity consumption with annual growth rates of 7% corresponding to a doubling per decade. Communication networks, computers, and data centers are equally responsible for the aforementioned energy consumption. The visionary goal of the Collaborative Research Center (CRC) HAEC (Highly Adaptive Energy-Efficient Computing) is to research technologies to enable computing systems with high energy-efficiency without compromising on high performance.Certainly, a straightforward way for improving energy efficiency is to reduce the energy consumption of every individual hardware component involved. However, it is equally important to understand how the software system can be adapted to the hardware and vice versa. Today, software is written without being energy-aware and is mapped agnostically onto generic hardware configurations of parallel machines. As the computational problems require a certain amount of problem-specific processing and intercommunication, a highly adaptive hardware system, which can optimize its configuration according to the needs of the software system, could generate a much higher level of efficiency. For this, system states of the application and hardware system need to be monitored and taken into account during run-time as well. In this regard, HAEC will concentrate on researching larger server systems, by innovating from applications to hardware.To achieve an integrated approach of highly adaptive energy-efficient computing (HAEC), the problem is approached at all levels of technology involved, the hardware, the computer architecture and operating system, the software modeling as well as the application modeling and runtime control levels. A novel concept is explored, the HAEC Box, involving optical and wireless chip-to-chip communication. This allows for a new level of run-time adaptivity, creating a platform for flexibly adapting the hardware configuration to the needs of the computing problem. To exploit this capability at the software level, a hierarchical energy-control loop infrastructure is researched, which is responsible for the control of the energy trade-off based on the current hardware state as well as the context-aware application requirements. Hence, the HAEC CRC is a first attempt to achieve high adaptivity and energy efficiency with an integrated approach going through all levels, from hardware components all the way to the application software.In Phase I, we broke the ground for the HAEC vision by coming up with significant progress in novel technologies as well as generating ideas for the realization of this vision. In Phase II we have picked up these ideas to show revolutionary approaches that drastically reduce the HAEC communications power consumpti