Escali Escali, LLC Network Router 4.4 User Manual
Scali MPI Connect Release 4.4 Users Guide
40
0: MPI_Bcast root: 0 Id: 0
my_count = 32768
0: MPI_Scatter Id: 1
1: MPI_Init
1: MPI_Comm_rank Rank: 1
1: MPI_Comm_size Size: 2
1: MPI_Bcast root: 0 Id: 0
my_count = 32768
1: MPI_Scatter Id: 1
1: MPI_Reduce Sum root: 0 Id: 2
1: MPI_Bcast root: 0 Id: 3
0: MPI_Reduce Sum root: 0 Id: 2
0: MPI_Bcast root: 0 Id: 3
1: MPI_Gather Id: 4
1: MPI_Keyval_free
0: MPI_Gather Id: 4
0: MPI_Keyval_free
my_count = 32768
0: MPI_Scatter Id: 1
1: MPI_Init
1: MPI_Comm_rank Rank: 1
1: MPI_Comm_size Size: 2
1: MPI_Bcast root: 0 Id: 0
my_count = 32768
1: MPI_Scatter Id: 1
1: MPI_Reduce Sum root: 0 Id: 2
1: MPI_Bcast root: 0 Id: 3
0: MPI_Reduce Sum root: 0 Id: 2
0: MPI_Bcast root: 0 Id: 3
1: MPI_Gather Id: 4
1: MPI_Keyval_free
0: MPI_Gather Id: 4
0: MPI_Keyval_free
If more information is needed the arguments to SCAMPI_TRACE can be enhanced to request
more information. The option “-f arg;timing” requests a list of the arguments given to each MPI
call, including message size ( useful information when evaluating interconnect performance).
more information. The option “-f arg;timing” requests a list of the arguments given to each MPI
call, including message size ( useful information when evaluating interconnect performance).
Example:
%SCAMPI_TRACE="-f arg;timing" mpimon ./kollektive-8 ./uf256-8.pgm -- n1 2
0: +-0.951585 s 951.6ms MPI_Init
0: +0.000104 s 3.2us MPI_Comm_rank Rank: 0
0: +0.000130 s 1.7us MPI_Comm_size Size: 2
0: +0.038491 s 66.3us MPI_Bcast root: 0 sz: 1 x 4 = 4 Id: 0
my_count = 32768
0: +0.038634 s 390.0us MPI_Scatter Id: 1
1: +-1.011783 s 1.0s MPI_Init
1: +0.000100 s 3.8us MPI_Comm_rank Rank: 1
1: +0.000129 s 1.7us MPI_Comm_size Size: 2
1: +0.000157 s 69.6us MPI_Bcast root: 0 sz: 1 x 4 = 4 Id: 0
my_count = 32768
1: +0.000300 s 118.7us MPI_Scatter Id: 1
0: +0.039267 s 38.8ms MPI_Reduce Sum root: 0 sz: 1 x 4 = 4 Id: 2
0: +0.078089 s 18.8us MPI_Bcast root: 0 sz: 1 x 8 = 8 Id: 3
1: +0.000641 s 56.2us MPI_Reduce Sum root: 0 sz: 1 x 4 = 4 Id: 2
1: +0.000726 s 113.6us MPI_Bcast root: 0 sz: 1 x 8 = 8 Id: 3
1: +0.002547 s 99.0us MPI_Gather Id: 4
0: +0.079834 s 579.5us MPI_Gather Id: 4
1: +0.002758 s 26.4us MPI_Keyval_free
0: +0.113244 s 1.4us MPI_Keyval_free
0: +0.000104 s 3.2us MPI_Comm_rank Rank: 0
0: +0.000130 s 1.7us MPI_Comm_size Size: 2
0: +0.038491 s 66.3us MPI_Bcast root: 0 sz: 1 x 4 = 4 Id: 0
my_count = 32768
0: +0.038634 s 390.0us MPI_Scatter Id: 1
1: +-1.011783 s 1.0s MPI_Init
1: +0.000100 s 3.8us MPI_Comm_rank Rank: 1
1: +0.000129 s 1.7us MPI_Comm_size Size: 2
1: +0.000157 s 69.6us MPI_Bcast root: 0 sz: 1 x 4 = 4 Id: 0
my_count = 32768
1: +0.000300 s 118.7us MPI_Scatter Id: 1
0: +0.039267 s 38.8ms MPI_Reduce Sum root: 0 sz: 1 x 4 = 4 Id: 2
0: +0.078089 s 18.8us MPI_Bcast root: 0 sz: 1 x 8 = 8 Id: 3
1: +0.000641 s 56.2us MPI_Reduce Sum root: 0 sz: 1 x 4 = 4 Id: 2
1: +0.000726 s 113.6us MPI_Bcast root: 0 sz: 1 x 8 = 8 Id: 3
1: +0.002547 s 99.0us MPI_Gather Id: 4
0: +0.079834 s 579.5us MPI_Gather Id: 4
1: +0.002758 s 26.4us MPI_Keyval_free
0: +0.113244 s 1.4us MPI_Keyval_free
There are a number of parameters for selecting only a subset, either by limiting the number of
calls and intervals as described above under ‘Timing’ , or selecting or excluding just some MPI
calls.
calls and intervals as described above under ‘Timing’ , or selecting or excluding just some MPI
calls.
4.2.2 Features
The "-b" option is useful when trying to pinpoint which MPI-call has been started but not
completed (i.e. are deadlocked). The "-s/-S/-c/-C" -options also offer useful support for an
application that runs well for a longer period and then stop, or for examining some part of the
execution of the application.
The "-b" option is useful when trying to pinpoint which MPI-call has been started but not
completed (i.e. are deadlocked). The "-s/-S/-c/-C" -options also offer useful support for an
application that runs well for a longer period and then stop, or for examining some part of the
execution of the application.