Navigation and service

Monitoring JUROPA-3

Site: Jülich Supercomputing Centre (JSC)
Batch System: Torque/Moab and SLURM
Peak-Performance: 34 Teraflops
Overall cores: 960 cores
Architecture: Test cluster, heterogenous


JUROPA-3 is a prototype cluster used for preparation of JUROPA's successor system. It represents a very heterogenous system with 60 compute nodes in total. Four nodes are enhanced with NVIDIA Tesla K20X GPUs and another four comprise Xeon Phi 5110P nodes. These special nodes are not yet distinguishable from the other nodes displayed with LLview. This system is divided into two partitions, each managed by a different batch system. The upper partition labeled with ZEA-1 is managed with Torque/Moab and the lower JSC partition is managed with SLURM. Therefore, two data gathering processes need to be triggered and later merged for this display, because each batch system is monitored on its own login node and requires different adapter scripts to be called. The increasing specificity, heterogeneity and size of the future systems are the challenges for a useful monitoring display. Continuous development and adaption of LLview to support new architectures and batch systems ensures that it can be applied to the next generation supercomputers.