i getting different time measurements every time run mpi application. measuring time this:
mpit1 = mpi_wtime(); // work mpit2 = mpi_wtime(); mpielapsed = mpit2 - mpit1; the behavior same, or without mpi_barrier(mpi_comm_world);. result of program, of course, correct every time executable runs.
check extreme difference got 5x5 matrix:
0.00025 seconds // first run 0.10728 seconds // second run moreover, problem occurring larger matrices, 2000x2000:
0.55996 seconds // first run 0.83657 seconds // second run am measuring time in wrong way? or knowing issue mpi? note time reported master node.
i running program in computer, have logged in ssh. recall getting faster execution when sitting in computer. user logged in:
gsamaras@pythagoras:~/konstantis/cholesky$ gsamaras internet connection gsamaras@pythagoras:~/konstantis/cholesky$ edit many runs:
1.6035 0.422967 1.53891 0.347107 0.761991 1.22824 0.934829 0.45618 0.794225 0.554099 1.09323 0.421858 0.621015 0.889867 0.613867 0.459144 2.02273 0.67689 2.2390 1.0448 0.751333 edit_2
- i got timings, calling program many times in run.sh.
- i running program in one machine, 4 cores.
- i using mpich2.
from colleague , patrick concluded that:
- the startup time mpi can vary lot.
- relatively small execution times not trust-worthy , may vary lot.
Comments
Post a Comment