JUNE 18–22, 2017
FRANKFURT AM MAIN, GERMANY
FRANKFURT AM MAIN, GERMANY
Presentation Details |
|||||
| Name: |  |
(PP25) Runtime Exploitation of Application Dynamism for Energy-efficient Exascale Computing | |||
| Time: | |
Tuesday, June 20, 2017 03:15 pm - 03:45 pm |
|||
| Room: | Booth #L-212 | ||||
| Breaks: | 03:15 pm - 03:45 pm Coffee Break | ||||
| Presenter: | Lubomir Riha, ZIH / TU Dresden | ||||
| Abstract: | The READEX project targets energy efficiency of HPC systems and is funded by the European Union's Horizon 2020 research and innovation programme under grant agreement No 671657.
The project partners come from different fields: HPC, embedded systems, academia and industry. The READEX tool suite increases the energy efficiency of HPC applications by employing different system configurations, called tuning parameters. One example of such a parameter is the CPU clock frequency. If parts of an application cannot enjoy high frequency, the frequency is reduced to conserve energy. Other parameters include application parallelism lowering, the selection of objective-dependent solvers and tuning of libraries.
The workflow of the READEX tool suite is as follows. First, the target application is instrumented so that different code regions become distinguishable. The result of the instrumentation is then used to perform early-stage trial runs. Here, single code regions and phases are grouped into scenarios. A scenario defines a set of optimal parameters for a given code region. Each scenario is stored in a tuning model. We call this process Design Time Analysis. When production runs of the target application is executed, the instrumentation is reused at runtime. Whenever a known scenario is detected, the tuning parameters are set according to the tuning model. Depending on the tuning objective, the effect may be a reduction in runtime or energy.
This poster provides an overview of the project, future developments and an initial analysis of the ESPRESO library's tuning potential. For more information, please visit http://www.readex.eu Zakaria Bendifallah, Intel Martin Beseda, IT4Innovations National Supercomputing Center, Technical University of Ostrava Othman Bouizi, Intel Anamika Chowdhury, Technische Universität München Kai Diethelm, Gesellschaft für numerische Simulation GmbH Michael Gerndt, Technische Universität München Andreas Gocht, Technische Universität Dresden, ZIH Daniel Hackenberg, ZIH, TU Dresden Venkatesh Kannan, Ireland's High-Performance Computing Centre
Per Gunnar Kjeldsberg, Norwegian University of Science and Technology Madhura Kumaraswamy, Technische Universität München Michael Lysaght, ICHEC (Irish Centre for High End Computing)
Umbreen Sabir Mian, Technische Universität Dresden Wolfgang E. Nagel, Technische Universität Dresden, Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH) Vojtech Nikl, Brno University of Technology Nico Reissmann, Norwegian University of Science and Technology
Lubomir Riha, IT4Innovations National Supercomputing Center, Technical University of Ostrava Robert Schöne, TU Dresden, ZIH Mohammed Sourouri, Norwegian University of Science and Technology Ondrej Vysocky, IT4Innovations National Supercomputing Center, Technical University of Ostrava Jan Zapletal, IT4Innovations National Supercomputing Center, Technical University of Ostrava
|
||||
| Download | PP25_Lubomir.pdf (1413 KB) |
||||
| |||||