Escali 4.4 Manual Do Utilizador

Página de 81
Section: 4.5 Using SMC's built-in CPU-usage functionality
Scali MPI Connect Release 4.4 Users Guide 
45
4.5 Using SMC's built-in CPU-usage functionality
Scali MPI Connect has the capability to report wall clock time, and user and system CPU time 
on all processes with a built-in CPU timing facility. To use SMC's built-in CPU-usage-timing it is 
necessary first to set the environment variable SCAMPI_CPU_USAGE. 
The information displayed is collected with the system-call "times"; see man-pages for more 
information. 
The output has two different blocks. The first block contains CPU-usage by the sub monitors on 
the different nodes. One line is printed for each sub monitor followed by a sum-line and an 
average-line. The second block consists of one line per process followed by a sum-line and an 
average-line. 
For example, to get the CPU usage when running the image enhancement program do:
user% SCAMPI_CPU_USAGE=1 mpirun -np 4 ./kollektive-8 ./uf256-8.pgm
This produces the following report:
                                             ---------- Own ----------- ------ Own+Children ------
Submonitor timing stat. in secs      Elapsed     User   System      Sum     User   System      Sum
Submonitor-1@r9                        2.970    0.000    0.000    0.000    0.090    0.030    0.120
Submonitor-2@r8                        3.250   -0.000    0.000   -0.000    0.060    0.040    0.100
Submonitor-3@r7                        3.180   -0.000   -0.000   -0.000    0.050    0.030    0.080
Submonitor-4@r6                        3.190    0.010    0.000    0.010    0.090    0.020    0.110
Total for submonitors                 12.590    0.010   -0.000    0.010    0.290    0.120    0.410
Average per submonitor                 3.147    0.003   -0.000    0.003    0.073    0.030    0.103
                                             ---------- Own -----------
Process timing stat. in secs         Elapsed     User   System      Sum
kollektive-8-0@r9                      0.080    0.070    0.030    0.100
kollektive-8-1@r8                      0.050    0.020    0.040    0.060
kollektive-8-2@r7                      0.050    0.020    0.030    0.050
kollektive-8-3@r6                      0.010    0.020    0.020    0.040
Sum for processes                      0.190    0.130    0.120    0.250
Average per process                    0.048    0.033    0.030    0.062
Elapsed is walltime used by user-process/submonitor
User is cpu-time used in user-process/submonitor
System is cpu-time used in system-calls
Sum is total cpu-time used by user-process/submonitor