Ovni

Ovni is a instrumentation toolkit written in portable C, to capture detailed runtime traces from multiple sources like HPC applications, OpenMP or OmpSs-2 runtimes, the Linux kernel and TAMPI or MPI libraries.

The information captured by ovni is then converted into visual Paraver traces for analysis using a trace emulator. The traces provide a detailed insight into what is happening in each CPU, so we can identify the source of problems and bottlenecks that affect the performance of highly parallel HPC applications.

The instrumentation overhead is reduced by using very short event sizes (typically 12 bytes) and a configurable level of detail, so only the information needed is recorded.

See more information at https://ovni.readthedocs.io

Performance & energy profiling of HPC/AI applications, regression testing in CI, scheduler research, post-mortem debugging of production failures, academic teaching of instrumentation.

Supercomputing centers building energy dashboards; Cloud-HPC providers offering performance-tuning services; Compiler/runtime vendors needing low-intrusion tracing; Semiconductor & system OEMs validating new hardware; University courses on performance engineering.