VAMPIR tracing tool on NSC's SGI

It provides an interactive graphical interface for the investigation of post mortem traces generated by MPI programs.

Users on the T3E compile their MPI programs with VAMPIR's tracing options enabled and link with routines to process this information. The trace file generated at runtime is viewed later on a workstation, where the user can interactively investigate the activity of processors and the messages being exchanged. Further information on its usage on the T3E is available here.

VAMPIR provides a highly detailed view of the message passing efficiency of an MPI code. It points out exactly those events on which the processor is waiting, and can, for instance, help you determine exactly why your program spends 40% of its time parked at an MPI_BARRIER or MPI_RECV.

The main differences between VAMPIR and other performance analysis tools are:

1. VAMPIR has a timeline with a display of the activities along this timeline. Apprentice and PAT are more oriented to detailed summaries of the activity of individual routines and can help determine whether or not excessive time is being spent in MPI_RECV or MPI_BARRIER or any user defined subroutine. Since Apprentice and PAT do not have a time line they can not however display when and how the time is distributed.

2. VAMPIR by default sums up all user defined subroutines into one lump sum. To get routine level display of anything else than the MPI routines the user has to insert library calls at the beginning and end of the routines of interest. This frequently increases the log files and memory usage to levels that require further intervention by the user to reduce the amount of tracing. Later this year there will be a conversion tool to convert PAT trace files to VAMPIR format which should simplify tracing of user defined subroutines.

See also the SGI FAQ file and a PostScript file.

For more information, mail faxen@nsc.liu.se.