The MPI library appears like any other library. Its memory behavior is assessed together with the rest of the application. This means that Freja looks at memory access patterns and complains about poor memory usage in relevant parts of the application, as well as in the MPI library.

On platforms where several MPI ranks are hosted within a single machine, the MPI library may use shared memory as a mechanism to transfer data between co-located ranks. The analysis of Freja does not extend into correctly classifying memory use from other processes. It considers each process in isolation.

Some tools analyze the communication between MPI peers, including judging communication speed and patterns, node imbalances or general effectiveness of the actual MPI-communication.

Such analysis is outside the scope of Freja, as it only looks at the behavior of each process in isolation.

On the other hand, different processes in a job may work on data with different characteristics, and therefore have different runtime behavior. In some cases, processes exercise completely different part of the application code. The reports produced by Freja for nodes with different behavior look different and point to optimizations that are important to their respective process.