Ptran Baseline Performance on Neptune
From WaterWiki
Contents |
[edit] Plot comparing baseline and profiling cases
[edit] Comments
The profiling overhead is quite high currently even after applying throttling during TAU profiling. Throttling is the processing of eliminating small functions which are called frequently from profiling. I fixed a few bugs in PETSc to get it to work with TAU. Still the integration of PETSc with TAU is suspect. I believe the fault lies within PETSc and not with TAU.
[edit] Base Ptran
Number of processors followed by time taken in each case.
1
CPU Time: 1.6489E-02 [sec] 2.7482E-04 [min] 4.5803E-06 [hr] Wall Clock Time: 1.6718E+02 [sec] 2.7863E+00 [min] 4.6439E-02 [hr]
2
CPU Time: 8.9710E-03 [sec] 1.4952E-04 [min] 2.4919E-06 [hr] Wall Clock Time: 9.5216E+01 [sec] 1.5869E+00 [min] 2.6449E-02 [hr]
3
CPU Time: 6.2020E-03 [sec] 1.0337E-04 [min] 1.7228E-06 [hr] Wall Clock Time: 7.1542E+01 [sec] 1.1924E+00 [min] 1.9873E-02 [hr]
[edit] Profiling with TAU
1
CPU Time: 3.9060E-02 [sec] 6.5100E-04 [min] 1.0850E-05 [hr] Wall Clock Time: 7.2067E+02 [sec] 1.2011E+01 [min] 2.0019E-01 [hr]
2
CPU Time: 7.3407E-02 [sec] 1.2235E-03 [min] 2.0391E-05 [hr] Wall Clock Time: 1.8561E+03 [sec] 3.0935E+01 [min] 5.1559E-01 [hr]
3
CPU Time: 7.5367E-02 [sec] 1.2561E-03 [min] 2.0935E-05 [hr] Wall Clock Time: 2.4203E+03 [sec] 4.0338E+01 [min] 6.7231E-01 [hr]
