Exploitation workshop
Workshop program
Tuesday, September 11
| 12:30 | Registration | Lunch session |
| Keynote | 13:00-14:00 | Natalie Bates: Towards a Roadmap for HPC Energy Efficiency |
| 14:00 | Coffee Break |
| eeClust: Session 1 | |
| 14:30-14:50 | Thomas Ludwig: Energy-Efficient Cluster Computing |
| 14:50-15:10 | Timo Minartz: Power Saving Potential |
| 15:10-15:30 | Daniel Molka: eeMark Design and VampirTrace Enhancements |
| 15:30 | Coffee Break | Afternoon session |
| eeClust: Session 2 | |
| 16:00-16:20 | Willi Homberg: Scalasca Enhancements in the eeClust Project |
| 16:20-16:40 | Stephan Krempel: eeDaemon Design and GridMonitor Enhancements |
| 16:40-17:00 | Timo Minartz: Evaluation |
| 17:00 | Coffee Break |
| Keynote | 17:30-18:30 | Enrique S. Quintana-Orti: Doing Nothing to Save Energy in Matrix Computations |
Invited Talks Abstracts
Natalie Bates
Towards a Roadmap for HPC Energy Efficiency
The growth rate in energy consumed by data centers in the United States has been declining in the past five years compared to its earlier accelerating pace. This reduced growth rate was achieved in large part due to energy efficiency improvements. Measuring, monitoring and managing usage has been key to making these improvements in energy efficiency. The metric Power Usage Effectiveness (PUE) has been effective in driving the energy efficiency of data centers, but it has limitations. PUE does not account for the power distribution and cooling losses inside the IT equipment, which is particularly problematic for HPC. Similarly, reporting performance and analyzing the amount of power used to run High Performance Linpack for a Top500 and/or Green500 submission has been successful in helping to drive improvements in supercomputing system energy efficiency. Power efficiency, (Megaflops per watt) shows average efficiency nearly tripling between 2007 and 2011. But just as PUE isn’t perfect for the data center, so are there problems with the power/energy measurement methodologies, workloads and metrics for supercomputer systems. Work is actively being done on both of these topics. Performance analysis capabilities and tools have been honed over the past decades and today we must develop the same capabilities and tools for energy and power analysis. The ability to achieve the 20MW target for an Exascale system is challenging and will require shifts in architecture, technology and application usage models as well as tighter coupling between the data center infrastructure and the computer system. This talk will describe current work and industry trends as well as layout a future roadmap for energy and power measurement capability, methodologies and metrics that will help us hit our 20 MW target.
Doing Nothing to Save Energy in Matrix Computations
Enrique S. Quintana-Orti
Power is becoming a crucial challenge that the high performance computing community will have to face to efficiently leverage the Exascale systems that will be available at the end of this decade. In this talk we will address several aspects related to this issue on multicore and many-core (GPU) processors. Specifically, we introduce a simple model of power dissipation, and a tools to compose a power tracing framework. Then, using experimental data, we evaluate both the potential and real energy reduction that can be attained for the task-parallel execution of dense and sparse linear algebra operations on multi-core and many-core processors, when idle periods are leveraged by promoting CPU cores to a power-saving C-state.