Turbulent Flow Simulation

Authors (in alphabetic order)

• Marten Lienen
• Peter Münch
• Walter Simson
• Josef Winter

Libraries

We require the following libraries

• openmpi version 1.10.1
• petsc version 3.5.4
• gtest version 1.6.0

Setup

For development you should install the following tools

• clang: We use clangs clang-format to keep our coding style consistent.
• cmake: We use cmake to generate a nicer Makefile than the one we got from the tutors.
• gtest: GTest is used for unit testing. Run the tests with ./tests.

# Run these once at the start of your session or put it into your
# .bashrc/.bash_profile. We do not want these values in our Makefile/
# CMakeLists.txt because they will be different for everyone and would thus
# generate lots of useless change sets.
export PETSC_DIR=<..>
export PETSC_ARCH=<..>

# Run this once in the beginning to generate a nice Makefile. You only need to
# run this again when you modified the CMakeLists.txt, for example when you add
# a new .cpp file.
cmake .

# Recompiling the program just needs a simple make. This make also has proper
# dependency tracking so that you should basically never need to use make clean.
make

Connecting to the cluster

# Read the usage information of the cluster connection script
scripts/cluster/connect -h

# Connect to cluster
# To connect to the linux cluster from lrz, pass `-c linux`
scripts/cluster/connect ...

# On the cluster

# Start by loading the git module
module load git

# Clone the repository
git clone https://github.com/cqql/turbulence

# Move into it
cd turbulence

# Load necessary modules (PETSc etc.)
#
# *Note the . at the beginning of the line!* This makes the script modify the
# environment of your current shell instead of creating a subshell for the
# execution of the script.
. scripts/cluster/load-modules

# Parallel Programming Analysis tools can also be loaded with
. scripts/cluster/load-modules --tuning

# Compile the simulation
cmake -DCLUSTER=ON -DCMAKE_BUILD_TYPE=Release .
make

# Submit a batch job to the cluster
scripts/cluster/run -e <you@tum.de> <OUTPUT DIR> ns scenario.xml

Profiling

There is also a PROFILING option in our CMake configuration to enable profiling that will produce a gmon.out that you can analyze with gprof.

The SCOREP option in CMake makes with the ScoreP instrumentation.

# Configure a profiling build
cmake -DPROFILING=ON .

# Alternatively configure with SCOREP. This can only be done on the Linux
# cluster after the proper modules have been loaded.
cmake -DSCOREP=ON -DCLUSTER=ON .

# Compile it
make

# Run it
./ns ...

# Analyze the results
gprof ./ns gmon.out

HDF5

We also support HDF5 as an alternative output format to VTK. To enable it, pass -DHDF5=on to cmake and enable it in your scenario with <hdf5 enabled="true" interval="0.1">result.h5</hdf5>.

Style Guide

We use google's C++ style guide. It is a reasonable guideline, used by a large organization to great success and at the same time it is an interesting read and can maybe even teach you something about C++.

You can use the script in scripts/format to automatically reformat all source files. Alternatively try one of the integrations of clang-format into various editors/IDEs.