Build Instructions

DXT Explorer requires a Darshan log file collected with tracing data. The Darshan eXtended Tracing (DXT) support is disabled by default in Darshan. To enable tracing globally for all files, you need to set the DXT_ENABLE_IO_TRACE environment variable as follows:

export DXT_ENABLE_IO_TRACE=1

To enable tracing for particular files you can refer to the Darshan’s documentation page.

Installing through git

Note

In Perlmutter (NERSC) you might need to load Darshan and Python modules if they are not already loaded. For other systems, please refer to their documentation to use the correct module name.

module load python
module load darshan

Note

In Summit at OLCF you need to follow this set of instructions.

module load python

conda create -n py310-dxt python=3.10
source activate py310-dxt
conda install arrow-cpp=10.0.1 pyarrow=10.0.1

git clone https://github.com/hpc-io/dxt-explorer
cd dxt-explorer

pip install .

dxt-explorer samples/YOUR-DARSHAN-FILE.darshan

conda deactivate

Run the below command to install some required Python libraries:

pip install -r requirements.txt

Then install dxt-explorer using the following command:

pip install .

Installing through pip

To install through pip, just run the following command:

pip install dxt-explorer

Warning

If you are installing dxt-explorer through pip, make sure the Darshan version installed on the machine matches the pyDarshan version installed through pip, otherwise you might get the following error:

darshan.discover_darshan.DarshanVersionError

Note

In NERSC systems (i.e., Cori or Perlmutter) you might need to load the Darshan module if it is not already loaded. For other systems, please refer to their documentation to use the correct module name.

module load darshan

Build with Spack

You can also use Spack to install dxt-explorer:

spack install dxt-explorer

Note

Use the following installation guide to install spack on your machine if it is not already installed: https://spack-tutorial.readthedocs.io/en/latest/tutorial_basics.html

Docker Image

You can also use a Docker image already pre-configured with all dependencies to run DXT Explorer:

docker pull hpcio/dxt-explorer

Since we need to provide an input file and access the generated .html files, make sure you are mounting your current directory in the container and removing the container after using it. You can pass the same arguments described above, after the container name (dxt-explorer).

docker run --rm --mount \
    type=bind,source="$(pwd)",target="/dxt-explorer" \
    dxt-explorer darshan/<FILE>.darshan