foRMA basic streaming version

This is the foRMA basic streaming version. Streaming because of the was profile information is calculated "on the fly" while the SST Dumpi traces are being parsed.

It offers reduced memory consumption in comparison to the foRMA baseline version and provides the basic metrics of an execution summary, i.e. operation duration analysis and summary, as well as data transfer volumes. These metrics are provided as an execution-wide summary, as a per-rank summary, but also as a per window-epoch summary.

foRMA baseline

This is the foRMA baseline version. It constructs an RMA view of an execution traced with the use of SST Dumpi and then uses this view in order to create an execution profile for RMA operation timings, data transfers, and synchronization characteristics.

The RMA view is constructed by extracting from the SST Dumpi trace, information relevant to RMA operations -- in this version, limited to the op codes MPI_Win_create, MPI_Win_free, MPI_Win_fence, MPI_Get, MPI_Put, and MPI_Accumulate. The information includes operation duration, operation starting time, op code, target rank, data transfer bytes.

The above data is organized hierarchically, indexed first by MPI rank, then by memory window ID, then by epoch number. This RMA view of the execution facilitates the calculation of metrics not only for one particular rank or aggregates over the entire execution, but also the comparison of particular metrics across ranks, e.g. for a particular window, a particular op code, or a particular epoch.