Navigation and service

Monitoring JUDGE

Site: Jülich Supercomputing Centre (JSC)
Batch System: Torque/Moab
Peak-Performance: 239 Teraflops
Overall cores: 2472 cores, 412 graphic processors
Architecture: GPU cluster, CPUs Intel Xeon X5650, GPUs NVIDIA Tesla M2050/M2070


JUDGE is a GPU cluster with 206 compute nodes with each node connected to two NVIDIA Tesla GPUs. Since TORQUE provides data on GPU usage per job, LLview can interpret this data and visualize the GPU usage among all running job. Each compute node is rendered with two large rectangles for the GPUs and 24 small rectangles for the CPUs. In addition, the overall GPU usage is indicated by the yellow line in the three-day-history at the bottom of the display. Note, that due to its architecture LLview can only gather status data available from a login or master node of the target system. In most cases batch system commands are called and parsed for the status data displayed with LLview. This ensures, that no additional daemon needs to be installed on the compute nodes and therefore guarantees the scalability of LLview as the status data is simply retrieved from the central management system of the supercomputer.