Navigation and service

Monitoring JUROPA

Site: Jülich Supercomputing Centre (JSC)
Batch System: Torque/Moab
Peak-Performance: 207 Teraflops
Overall cores: 26304 cores
Architecture: Sun/Bull cluster using Intel Xeon X5570

LLview JUROPA

The general purpose cluster JUROPA is maintained by JSC. A cluster is usually presented by LLview with one rectangle per core in the nodes view. For a clearer view of large clusters like JUROPA a node can display usage bars showing the percentage of cores used by each job. This allows to see the amount of compute resources used by each job without the need for displaying one rectangle per core. Moreover, this display is a good example for the flexibility given with LLview's layout configuration. The nodes view is visually separated into the two logical partitions of the vendors Sun and Bull. Nodes can be placed into racks, which again are arranged in a configurable grid layout. Images can be attached to each rack or node, which helps to better understand the system architecture.

The scheduler simulator JuFo is used for predicting the future schedule of JUROPA. With the help of JuFo's test framework the actual scheduling algorithm is closely reconstructed.

For demo purposes this display can be split into multiple LLview instances, where each of them shows a specific part of the system status. For JUROPA it is useful to provide one display presenting only the nodes view, while the other shows a larger version of the prediction. Thus, multiple layout configurations for LLview help to serve the needs of the different user groups such as system administrators, users or visitors.