Each rank contacts the license server separately, which for wide jobs could be a bottleneck or a hard limitation.

Burst mode may not be effective to reduce overhead if ranks communicate with each other.

Not all MPI versions make the rank number available as an environment variable. This means that Freja will not be able to produce sample files with names that directly relate to the MPI rank number. Instead, some sequential numbering scheme is used.

Some MPI environments (e.g., Cray) use optimizations to stage binaries before executing them. Such optimizations need to be disabled when sampling.

Automatic processor detection at the time of report generation looks at the current processor in the machine where the report generation is performed. In a heterogenous system that processor type does not need to be identical to the processors in the compute nodes.