IPM: MPI performance profiling

IPM is a tool which gives rapidly an overview over the time spent in the different MPI calls. It is very simple to use and can also give a html output with graphical representation of the results.

In your script or on the command line just  write

module load ipm
ipm_start

and run you application as usual.

At the end of the run you will get an overview (standard output):

##IPMv0.982####################################################################
# 
# command : a.out  (completed)
# host    : stallo-1/x86_64_Linux          mpi_tasks : 1 on 1 nodes
# start   : 04/30/10/13:09:19              wallclock : 0.002069 sec
# stop    : 04/30/10/13:09:19              %comm     : 0.02 
# gbytes  : 8.27026e-02 total              gflop/sec : 0.00000e+00 total
#
##############################################################################
# region  : *       [ntasks] =      1
#
#                           [total]         <avg>           min           max 
# entries                          1             1             1             1
# wallclock                 0.002069      0.002069      0.002069      0.002069
# user                      0.011998      0.011998      0.011998      0.011998
# system                    0.024996      0.024996      0.024996      0.024996
# mpi                    3.96278e-07   3.96278e-07   3.96278e-07   3.96278e-07
# %comm                                  0.0191531     0.0191531     0.0191531
# gflop/sec                        0             0             0             0
# gbytes                   0.0827026     0.0827026     0.0827026     0.0827026
#
#
#                            [time]       [calls]        <%mpi>      <%wall>
# MPI_Comm_size          2.31201e-07             1         58.34         0.01
# MPI_Comm_rank          1.65077e-07             1         41.66         0.01
###############################################################################

(You can stop ipm with ipm_stop)

This will also produce a file like.

 MyName.1272624855.201741.0

You can then run the command (on the front-end, stallo-1 or stallo-2, not a compute-node):

ipm_parse -html MyName.1272624855.201741.0

Which produce a new directory with html files that you can visualize in your browser:

 firefox a.out_1_MyName.1272624855.201741.0_ipm_unknown/index.html

Note that the use of hardware performance counter are not implemeted yet on Stallo. Therefore IPM cannot give information about floating point operations, cache use etc.

For more details refer to:

http://ipm-hpc.sourceforge.net/userguide.html