JUQUEEN Extreme Scaling Workshop 2016

Feedback from last year's very successful workshop motivated the organization of a three-day workshop in February, during which the entire 28-rack JUQUEEN BlueGene/Q system was reserved for over 50 hours. Eight code teams were selected to use this opportunity to investigate and improve their application scalability, assisted by JSC Simulation Laboratory and Cross-sectional Team staff.

The applications Code_Saturne from Daresbury Laboratory and Seven-League Hydro from HITS (Heidelberg) both displayed strong scalability to 28 racks and thereby become candidates for High-Q Club membership. Existing members, CIAO from RWTH-ITV and iFETI from University Cologne and TU Freiberg, showed that they had additional solvers which also scaled acceptably, and in situ visualization was demonstrated with a CIAO/JUSITU/VisIt simulation running on 28 racks. Two adaptive mesh refinement libraries, p4est from University Bonn and IciMesh from EC Nantes, showed that they could respectively scale to 917,504 and 458,752 MPI ranks, but both encountered problems loading large meshes. Parallel file I/O limitations also prevented large-scale executions of the IEK-6/Amphos 21 PFLOTRAN subsurface flow and reactive transport code; however, an HDF5 import module for NEST developed by the EPFL Blue Brain Project could be optimized to use collective MPI file reading calls and enable large-scale data-driven neuronal network simulations.

The optimizations achieved with all codes will be a good basis for running bigger jobs on JUQUEEN in current and future compute time projects. Detailed results for each code provided by the participating application teams appeared in the JSC report FZJ-JSC-IB-2016-01, see http://juser.fz-juelich.de/record/283461

(Contact: Dr. Brian Wylie, b.wylie@fz-juelich.de)

JSC News, 18 March 2016