MAXI - Multi-system Application Extremescaling Imperative
JSC organized the mini-symposium MAXI from 3-4 September 2015 as part of this year’s International Conference on Parallel Computing (ParCo) 2015 in Edinburgh. The goal was to encourage users to report about scaling their applications on more than one HPC system. Various contributions from fields such as engineering, neuroscience, and earth modelling were presented. The researchers explained their codes and results gained on systems such as Tianhe-2 (TOP500 ranking #1), large Blue Gene/Q installations (#3, #5, and #9), K computer (#4), SuperMUC Phase 1 and Phase 2 (#20 and #21) and other x86 clusters, as well as the latest NEC vector architecture, sometimes even comparing their performance between architectures. Specific node level performance improvements were also presented as well as an update on the 2013 Gordon Bell award winning code. A broader spectrum of applications and system experience was covered by talks on scaling workshops and usage statistics from JUQUEEN, SuperMUC-2, and K computer. Those activities were reported to lead to easy transitions to even bigger scales, to help co-design future HPC systems, and enable production runs on the 28 racks of JUQUEEN or even simulations of bigger problem sizes.
Overall, the mini-symposium was well received and it became clear that more discussion and reports are desirable also covering problems encountered when using larger and larger systems. Not only should outstanding results be recognized but also the efforts required to reach them. Further information is available at http://www.fz-juelich.de/ias/jsc/MAXI.
(Contact: Dr. Dirk Brömmel, email@example.com)
JSC News, 21 September 2015