The figure shows ClusterCockpit measurement results from job that does regular filesystem operations. These lead to increased network and filesystem utilization. The plots show a regular pattern in the application behavior.
The figure shows ClusterCockpit measurements for a job that executes on multiple nodes in shared-mode. Three metrics are shown: node-level CPU load, averaged node-level CPU core load and averaged node-level CPU core time. While the CPU load metric may be influenced by other jobs running on the shared nodes, the CPU core load and CPU core time metrics use core-level measurements. These metrics show a difference in CPU utilization on different nodes.
The image shows ClusterCockpit measurement results for an application that is executed with multiple scaling configurations. It is executed with four different configurations on 1 core, 128 cores (1 node), 256 cores (2 nodes) and 512 cores (4 nodes). Three metrics are show for every case: Flops per core, memory bandwidth per socket and transmitted network packets per node. The plots show how with increased number of involved cores and nodes the performance degrades and the communication overh...
