- Johannes Gutenberg University of Mainz
- Fraunhofer Society Sankt Augustin
- University of Cologne
- ParTec Munich
- Aachen University
- Munich University of Technology
FaST stands for “Find a Suitable Topology for Exascale Applications” and is a research project exploring the temporal and spatial placement of processes in the supercomputers of the future. Based on the assumption that CPU performance will continue to increase at a faster rate than I/O performance, scheduling strategies are to be developed to monitor the relevant system resources and make local adjustments to the distribution of jobs. To ensure the efficient use of computing power, applications with different requirements need to be grouped intelligently. FaST allows for such a requirement-based assignment and optimizes this through the use of new scheduling algorithms. MEGWARE is involved in the project, using its in-house products to help create an infrastructure for measuring energy. We are integrating the results in our in-house cluster management software and helping evaluate them. The efficient use of computer technology is a major focus of our development work, since in the case of very large cluster systems, the technology currently available is reaching the limit of its capability. A more targeted use, based on the results of this project, will facilitate greater computing power.
Leading industry and research representatives from the field of High Performance Computing (HPC) have joined forces to establish the European Technology Platform for HPC (ETP4HPC). Among the founders of ETP4HPC are some of Europe’s leading high performance computing and supercomputing centers. In addition to the Jülich Supercomputing Centre, the Barcelona Supercomputing Center, the French Alternative Energies and Atomic Energy Commission (CEA), Cineca, the Fraunhofer Society, and the Leibniz Supercomputing Center (LRZ) are also involved in the European forum. MEGWARE is one of a total of fifteen members leading the technology platform committee.
The objective of this group is to produce competitive HPC solutions in Europe and to promote the development of technology. The aim is therefore to forge closer links in the future between research, development, and innovation in the field of HPC. Together with its partners, the ETP4HPC forum believes that in the coming years it will be at the forefront of the HPC industry, whose developments are anticipated to have an impact on research and technology. To this end, the targeted promotion and development of research programs, startups, and SMEs, as well as synergy effects with related information technologies, for example, are required.
ETP4HPC is working closely with the HPC infrastructure for the scientific community and industry in Europe, “Partnership for Advanced Computing in Europe” (PRACE), which controls the provision of computing capacities and the development of software and applications for science and industry at the European level.
- Chemnitz University of Technology
Since increasing computing power also brings with it higher energy requirements, energy-reducing measures need to be taken. The energy and power consumption of a given cluster’s subsystems can be optimized using specific control methods, procedures, and algorithms. However, the configuration parameters of the different subsystems also need to be set appropriately, whereby a predetermined balance between energy savings and potential performance loss needs to be considered. Since, in terms of their effect, the subsystems are not always autonomous, it is extremely hard to estimate the overall effect that can be achieved due to the complexity.
This is where the energie-POKER project comes in. This project was launched to support the step-by-step process of optimizing a cluster system in terms of its energy requirement through the systematic optimization of energy-relevant configuration data.
The project consists of two subprojects. Professors of Computer Architecture at Chemnitz University of Technology are working on the “EPOKER” subproject, in which a cluster energy configuration data manager is to be developed to consistently record, manage, and control all energy-relevant data. MEGWARE is managing the second subproject: “KNOW-2 Pow.” This project involves the realization of effective interaction with the cluster energy measurement system (“KNOWPow”), which we are developing as a supplementary system to the measurement systems available on the market to record additional performance and energy data.
We believe that the innovative concepts of the project will lead to success. Should it prove possible to successfully apply the chosen solutions to real-life cluster procurement processes, we intend to further develop the framework to a prototypical application in an in-house Benchmark Center.
- Bayer Business Service GmbH
- German Aerospace Center (DLR)
- Berlin University of Technology
- VOICE – Verband der IT-Anwender e.V. (Association of IT users)
GRIP is short for “Green IT Procurement” and describes a user-oriented test procedure for the procurement of energy-efficient IT components for data centers. In view of more powerful and energy-intensive infrastructures, it is especially important to develop and produce energy-efficient HPC systems for data centers.
Numerous studies confirm the ever-higher energy requirements of data centers. To improve energy efficiency, energy-saving hardware, dimensioned in line with its intended purpose, needs to be used. Energy-efficient components are, however, difficult to identify, since it is hardly ever possible to directly compare hardware from different manufacturers.
All parties involved in the project have therefore set themselves the goal of developing a test procedure that can be used to facilitate the identification of more energy-efficient, optimally dimensioned hardware. Based on a neutral definition of real-life application scenarios and typical load profiles of servers, we are developing standardized test procedures to make it possible to commit manufacturers to disclosing informative energy efficiency data. Until now, there has been no application-related, standardized test procedure for the quantitative rating of the energy efficiency of servers.