JUNE 19–23, 2016
FRANKFURT AM MAIN, GERMANY

Session Details

 
Name: BoF 11: Programming Models for Exascale Supercomputers – A Slow Transition or Complete Disruption?
 
Time: Tuesday, June 21, 2016
03:15 pm - 04:15 pm
 
Room:   Booth #220  
 
Breaks:02:45 pm - 03:30 pm Coffee Break
 
Speaker:   Valeria Bartsch, Fraunhofer ITWM
  Mark Bull, EPCC
  Jesús Labarta, BSC
 
Abstract:   Without knowing exactly what future HPC architectures will look like, it is not easy to define the next-generation programming models to go with them. Nevertheless, this is a problem that we need to start tackling now, to ensure applications are ready to exploit the potential of Exascale computing. In this BoF session we will address three important aspects of Exascale programming models. In the first talk we will set the stage, and investigate if an Exascale programming model can be achieved with a smooth transition from current approaches or whether a new (and disruptive) approach is needed. In the second talk, we will consider the question of how best to exploit heterogeneous architectures, and how to cope with system failures. In the third talk we will discuss if current programming models are enough and share hands-on experience from research to compare MPI and GASPI applications on the way to Exascale. The underlying and prevalent question in all of these talks will be: Is preserving the investment in legacy applications and opting for an evolution of existing models a viable approach? Or do we need a more drastic approach to fully exploit future architectures? The speakers and organisers are part of a collaborative effort of EU-funded research projects, who have undertaken extensive research in the field of future programming models: DEEP/-ER, EPiGRAM, EXA2CT, INTERTWinE, Mont-Blanc and Numexas. The talks will be inspired by and based on their findings and results.

Targeted Audience
This topic is of interest to those working with programming models – especially those researching the future of programming models. However, considering the co-design aspect that is of utmost importance for building next-generation supercomputers, this topic will also be interesting to HPC application developers.
We aim to attract young researchers in the HPC community – both from software and application development. They are especially encouraged to contribute to this already quite controversial discussion on future programming models.