Speaker:

Bronis R. de Supinski, Lawrence Livermore National Laboratory, USA

Contents:

Exascale systems are currently projected to exist by 2019. The architectures of these systems, which will employ unprecedented levels of parallelism, will present numerous challenges. Many of these challenges are already becoming common in current systems but will become even more significant. For example, multicore chips not only imply increased parallelism but also bring reduced memory size per core and memory bandwidth per core due to cost considerations. Similarly, many exascale systems will feature accelerators, such as GPUs present in the Chinese Tianhe-1A system, the current number two on the Top500 list. Other challenges, such as ensuring resilience in the face of soft errors and debugging large-scale parallel jobs, have existed for several years but could become insurmountable due to the architectural changes of exascale systems. To overcome these challenges, programming models for large-scale systems and the algorithms implemented in those models must change significantly. Further, we must adapt our system software and tools and their paradigms to address the unprecedented concurrency and other architectural trends. In this talk, I will outline the overall strategy of Lawrence Livermore National Laboratory to address these challenges and detail some recent advances that contribute to that strategy.

Date:

Wednesday, 5 October 2011, 13:30

Venue:

Jülich Supercomputing Centre, Hörsaal, building 16.3, room 006

Announcement as pdf: