Electronic poster: eeclust: energy-efficient cluster computing

The eeClust project aims at reducing the energy consumption of applications on a HPC cluster by an integrated approach of analysis, efficient management of hardware power-states and monitoring of the clusters power consumption. The application is traced and the trace file is analyzed - manually with Vampir and automatically with Scalasca - to determine phases in the application with non-optimal hardware utilization. The source-code is then instrumented with API calls to control a daemon which switches hardware power-states at runtime. This daemon is aware of shared resources (e.g. the network interface) and only switches a resource to a lower power-state when all processes sharing that resource do not need it. The ParaStation Grid Monitor is used to monitor and visualize the power consumption and hardware usage of the cluster. This poster gives an overview of the project and presents the analysis, hardware management and monitoring aspects in more detail.

[1]  Bernd Mohr,et al.  eeClust: Energy-Efficient Cluster Computing , 2010, CHPC.

[2]  Bernd Mohr,et al.  Managing hardware power saving modes for high performance computing , 2011, 2011 International Green Computing Conference and Workshops.

[3]  Wolfgang E. Nagel,et al.  Flexible workload generation for HPC cluster efficiency benchmarking , 2012, Computer Science - Research and Development.

[4]  Bernd Mohr,et al.  Determine energy-saving potential in wait-states of large-scale parallel programs , 2011, Computer Science - Research and Development.

[5]  Thomas Ilsche,et al.  The VampirTrace Plugin Counter Interface: Introduction and Examples , 2010, Euro-Par Workshops.