JUQUEEN Extreme Scaling Workshop 2017
From 23 to 25 January 2017, JSC organized its eighth IBM Blue Gene Extreme Scaling Workshop. The entire 28-rack JUQUEEN Blue Gene/Q was reserved for over 50 hours to allow six selected code teams to investigate and improve the scalability of their applications. Ultimately, all six codes managed to run using the full complement of 458,752 cores, most with over 1.8 million threads.
MPAS-A (from KIT / NCAR) and the pe rigid body physics engine (University of Erlangen) were both able to display strong scalability to 28 racks and thereby become candidates for High-Q Club membership. After participating in the 2015 workshop, MPAS-A returned with a higher resolution dataset and substantially improved file I/O using SIONlib to successfully manage its largest ever global atmospheric simulation. While ParFlow (University of Bonn/IBG at Jülich) had recently demonstrated execution scaling to the full JUQUEEN without file writing enabled, during the workshop the focus was on investigating file I/O performance, which remains a bottleneck. High-Q Club member KKRnano (IAS at Jülich) investigated the scalability of a new solver algorithm developed to handle a million atoms, while the latest version of CPMD was tested with a large 1500-atom system. Both of these quantum materials codes revealed performance limitations at larger scales. The final code was a prototype multi-compartmental neuronal network simulator, NestMC (JSC SimLab Neuroscience), which compared scalability of different threading implementations.
Detailed results for each code provided by the participating application teams were published in the JSC technical report FZJ-JSC-IB-2017-01, see http://juser.fz-juelich.de/record/828084.
(Contact: Dr. Brian Wylie, email@example.com)
JSC News No. 248, March 2017