Skip to content


Repository files navigation


This repo is associated with experiments done on NCSA's Blue Waters system to improve our understanding of computational uncertainty and reproducibility.

  • Due to the non-deterministic nature of this experimental procedure, we expect scientists attempting to replicate this work will only replicate to one or two significant digits for each experiment. These experiments were carried out during the later half of 2016 and so a researcher attempting to replicate this work on Blue Waters may find specific versions of compilers we used (and even Blue Waters itself) unavailable in the future.
  • gif displaying experimental output

Software Summary

We present here a list of software which was used to perform the cosmology simulations along with the software to analyze the output.

Cosmology Simulation

As a summary, we quickly go over the software associated with the cosmology simulation itself. We use The following libraries and frameworks either explicitly or as a dependency.

  • Enzo 2.5

  • HDF5 1.8 (We tested with 1.8.16 and 1.8.20)

  • HDF5 must also have shared libraries and mpi compatibility enabled.

  • MPI implementation

  • We had trouble building our software stack with openmpi. In the end, we went with MPICH 3.2 or a MPICH 3 compiled for the machine in question.

  • zlib 1.2.11 (Other 1.2 versions of zlib should be okay too.

The Spack package spec for enzo on a recent Arch Linux machine is:

spack find -d -v enzo
==> 1 installed packages.
-- linux-archrolling-x86_64 / gcc@8.1.1 -------------------------
    enzo@2.5~bitwise~bluewaters~cray mode=debug patches=c473271a7c7d14f5ca71c1be58a7d78c71156282e7d62519861782cc8100eb9c,fac46ec62d9dcc4aaf3808c39d7ce7f4a84d983e18f2d0f5f52f129ef5a42380
            ^mpich@3.2.1 device=ch3 +hydra netmod=tcp +pmi+romio~verbs

Analysis Toolchain

  • Yt 3.4.1 (We also used 3.3.5)
  • libconfig 1.5
  • Python libconf 1.0.1

The Spack package spec for the various toolchain libraries is:

spack find -d -v py-yt                                                                                  
==> 1 installed packages.
-- linux-archrolling-x86_64 / gcc@8.1.1 -------------------------
                ^python@2.7.15+dbm~optimizations patches=123082ab3483ded78e86d7c809e98a804b3465b4683c96bd79a2fd799f572244 +pic+pythoncmd+shared~tk+ucs4
                ^hdf5@1.10.2~cxx~debug~fortran+hl+mpi patches=57cee5ff1992b4098eda079815c36fc2da9b10e00a9056df054f2384c4fc7523 +pic+shared~szip~threadsafe
                    ^openmpi@3.1.1~cuda+cxx_exceptions fabrics= ~java~memchecker~pmi schedulers= ~sqlite3~thread_multiple+vt
                            ^numactl@2.0.11 patches=592f30f7f5f757dfc239ad0ffd39a9a048487ad803c26b419e0f96b8cda08c1a
                    ^openblas@0.3.2 cpu_target= ~ilp64 patches=47cfa7a952ac7b2e4632c73ae199d69fb54490627b66a62c681e21019c4ddc9d +pic+shared threads=none ~virtual_machine
                ^qhull@2015.2 build_type=RelWithDebInfo patches=10ddc62a1600b0ede23fb0502e29dca389d18689fd8dfe7a849e3a989c0e607e
        ^rockstar@yt~hdf5 patches=4a7edd21e5798c454203c75afa51e699ef20786b6119d2984cd7d805a3f9ef3d,b6adb7547f156d6b07f3ba649f8554c9cfdcdad8109aeacbe8b6f8aaced1b797
spack find -d -v libconfig 
==> 1 installed packages.
-- linux-archrolling-x86_64 / gcc@8.1.1 -------------------------
spack find -d -v py-libconf
==> 1 installed packages.
-- linux-archrolling-x86_64 / gcc@8.1.1 -------------------------
        ^python@2.7.15+dbm~optimizations patches=123082ab3483ded78e86d7c809e98a804b3465b4683c96bd79a2fd799f572244 +pic+pythoncmd+shared~tk+ucs4

Build procedure

Manually managed

We are confident that if the user builds all the listed packages above and sets their PATH, LD_LIBRARY_PATH, PKG_CONFIG_PATH, and PYTHONPATH appropriately with respect to the installation directory, they will be able to run the initial cosmology simulation as well as the analysis which follows.

Spack Managed

We provide a script called which uses spack to install the necessary packages. If spack has trouble installing a given package, you can try adjusting the version number, as well as applying a patch to fix your problem.

Run Procedure

Provided with this software package is the directory 32-7-10Mpc-z0. This directory contains all the necessary parameter files and initial conditions files to run enzo.

Once Enzo is built and the binary enzo runs, navigate to the 32-7-10Mpc-z0 directory, and execute mpirun -np ${NUM_CORES} enzo gas_plus_dm_amr_multiphys.enzo

Reduced test case

In order to quickly test that you have this simulation running correctly, We provide a reduced test case which should run in < 1 hour for most users.

16-3-10Mpc-z0 contains the already generated initial conditions and the enzo parameters definitions file.


This repo is associated with experiments done on NCSA's Blue Waters system to improve our understanding of computational uncertainty and reproducibility







No releases published


No packages published