Maintainer: Mario Bielert, ZIH Tools Team @ TU Dresden

Collecting measurement data from HPC applications can be difficult. Lightweight Node-Level Performance Monitoring (Lo2s, IPA: [ˈloːtʊs]) presents with its system monitoring a novel viewpoint on application behavior, which can be used to find performance bottlenecks on a node-level. In order to use this measurement system effectively, users must be aware of both the common performance-engineering issues that affect all performance measurements and the specifics of how to use the two measurement modes and various additional metric recordings of Lo2s to collect accurate and relevant performance data.

Students should learn how to use Lo2s to produce sampling data for their HPC applications and compute nodes. They should understand how to evaluate and improve the quality of their measurement. They should understand at least some of the relevant specialized ways to configure and use Lo2s that apply to a problem of immediate interest to them.

Able to sample applications including one or more parallel paradigms from the following:

• MPI (node-level)
• OpenMP
• Pthreads

Able to sample applications and nodes including at least one specialized form of measurement from the following:

• I/O performance
• Hardware counters
• Perf metrics
• Linux tracepoints

Able to use Lo2s in application monitoring and system monitoring mode