README

These are the release notes for opm-benchmarks

Opm-benchmarks currently consists of one benchmark with two included data sets.
It is the mpi version of the upscale_relperm code from opm-upscaling, and can
distribute to a maximum of 21 cpus in it's current form. The benhcmark is
copyrighted by Statoil, and licensed under GPL 3+. See COPYING for details.

ON WHAT PLATFORMS DOES IT RUN?
It is only developed for GNU/Linux. All the larger distributions are
usually tested.

HOW MUCH MEMORY DO I NEED?
The benchmark itself uses just over 100MB per process on the smallest data set.
Compiling the benchmark is quite memory intensive. Plan for at least one gigabyte
of available memory for each thread when compiling. Number of threads is given
with the -j option when calling make below.

BUILDING AND RUNNING THE BENCHMARK
Install dependent DUNE core modules from Debian Packages:

  sudo add-apt-repository -y ppa:opm/ppa
  sudo apt-get update
  sudo apt-get install libdune-common-dev libdune-geometry-dev libdune-grid-dev libdune-istl-dev

Then build the OPM modules from source:

  git clone https://github.com/OPM/opm-core.git -b release/2013.03
  git clone https://github.com/OPM/opm-porsol.git -b release/2013.03
  git clone https://github.com/OPM/opm-upscaling.git -b release/2013.03
  git clone https://github.com/OPM/dune-cornerpoint.git -b release/2013.03
  git clone https://github.com/OPM/opm-benchmarks.git

  modules='opm-core dune-cornerpoint opm-porsol opm-upscaling opm-benchmarks'

  for module in ${modules}; do
      mkdir ${module}/build
      cd ${module}/build
      cmake -DUSE_MPI=ON ../
      nice make -j 4
      cd ../../
  done

To run the benchmark, issue

  nice mpirun -np 4 ./opm-benchmarks/build/bin/upscale_relperm_benchmark

Change the number of threads by configuring the option -np.