Dies sind die archivierten Webseiten des Lehrstuhls für Programmierung und Softwaretechnik (PST).
Die Seiten des Software and Computational Systems Lab (SoSy) finden Sie auf https://www.sosy-lab.org/.

Informatik-Kolloquium Mi, 16.05.2012, 14:15 Uhr

— abgelegt unter:

Dr. Martin Schulz, Lawrence Livermore National Laboratory: Power and Energy Efficient High Performance Computing - A Challenge for the Entire System Stack

  • Kolloquium
Wann 16.05.2012
von 14:15 bis 15:45
Wo Raum 123, Oettingenstr. 67
Termin übernehmen vCal


Power consumption is emerging as the primary concern for exascale
computing. Both facility and cost concerns are turning energy
consumption into a hard constraint. This has significant consequences
for system architecture, system software and applications and requires a
concentrated effort across the entire exascale stack. While
traditionally most efforts have focused on transparent hardware
solutions, all system software layers and components and even then the
application itself can and must contribute to energy savings. In this
talk I will highlight a series of projects that aim at characterizing
the potential for energy savings through adaptive runtime systems,
exploiting slack caused by message delay and dynamically throttling
multithreaded and hybrid workloads. In addition, I will explore the
effects of hard power constraints on applications as well as initial
ideas on how applications can help contribute to power savings.


Martin is a Computer Scientist at the Center for Applied Scientific
Computing (CASC) at Lawrence Livermore National Laboratory (LLNL). He
earned his Doctorate in Computer Science in 2001 from the Technische
Universität München (Munich, Germany) and also holds a Master of Science
in Computer Science from the University of Illinois at Urbana Champaign.
He has published over 100 peer-reviewed scientific conference and
journal articles. He is a member of LLNL's ASC CSSE ADEPT (Application
Development Environment and Performance Team), the PI for the ASC/CCE
project on Open|SpeedShop, and the LLNL PI for the OASCR PetaTools
project on "Building a Community Tool Infrastructure around
Open|SpeedShop". He is further involved in the ASCR Co-Design Centers
CESAR (as LLNL PI) and ExMatEx (as task lead). Martin's research
interests include performance monitoring, modeling and analysis; memory
system optimization; parallel programming models and paradigms; tool
support for parallel programming; power efficiency for parallel systems;
and fault tolerance at the application and system level.