DASH Example bench.11.hdf-io

Examples Index


Benchmark for measuring the IO performance of DASH and HDF5 by storing a Matrix in an hdf5 file.

Important Execute this benchmark on a parallel file system like GPFS as the performance will decrease otherwise. This Benchmark requires a build with parallel HDF5 support.

Usage (DART-MPI)

$ DASH_MAX_UNIT_THREADS=<T> mpirun -n <P> ./bin/bench.11.hdf-io


Parameter Description Default
\(\texttt{T}\) Number of threads per unit (process) total CPU cores / \(P\)
\(\texttt{P}\) Number of units (processes) -
\(\texttt{sb}\) initial matrix size (sb x sb) 57344
\(\texttt{verify}\) verify read data (time consuming) 0

Sample Output

SuperMUC Haswell: 28 units, 1 thread / unit

Executed in the $WORK dir, which is mounted on GPFS.

$ mpiexec -n 28 ./bin/bench.11.hdf-io -sb 57344

Jobfile: bench.11.hdf-io.pbs

-- Runtime arguments
--   -sb          57344                              initial matrix size
--   -verify          0                                     verification
units, mpi.impl,     mb.unit,   mb.global,      init.s,     write.s,      read.s,  write.mb/s,   read.mb/s,      time.s
   28,    mpich,      896.00,    25088.00,        0.85,       27.95,       22.06,      897.49,     1137.36,       50.93