Source codes and simple test cases.
We will clone the repo on TRUBA and dispatch a SLURM job that will compile and execute the fiber counting code on the sample example_tensors/t0.tns tensor on a single node using 10 cores.
First, while logged into TRUBA, we clone the repository and navigate to it:
git clone https://github.com/SU-HPC/Computhon2022-1
cd Computhon2022-1
Then, we dispatch the job by executing the following command with <account> being our account on TRUBA:
sbatch example_scripts/script_1node_10core_1gpu.slurm -A <account>
The outputs of the job will be in the file fibers-%j.out and fibers-%j.err where %j is the job ID.
You can start sending jobs to the cluster directly using one of the given SLURM scripts in the directory example_scripts. You can either use the script example_scripts/slurm_template.slurm as a template and fill out the options you need (advanced), or use one of the ready made SLURM scripts that cover the three categories of the competition:
example_scripts/script_1node_10core_1gpu.slurm: this script will reserve a job that will use 1 node (server), 10 cores, 1 GPU, and a single task (process).example_scripts/script_1node_10core_4gpu.slurm: reserves a job that will use 1 node, 1 process (task), 10 cores, and 4 GPUs.example_scripts/script_3node_60core_12gpu.slurm: reserves a job that will use 3 nodes, 3 processes (tasks), 20 cores per process (60 cores total), and 4 GPUs per node (12 total).
The above three scripts include a paragraph at the top explaining the different options used to establish the job parameters. In addition, all the scripts will load the required modules to use OMP, CUDA, and MPU. Finally, the scripts, by default, will compile and execute the sample fiber counting code (fibers.cc) on the graph example_tensors/t0.tns.
To execute any of the above scripts (as well as the template script) use the following command:
sbatch example_scripts/<script> -A <account>
where <script> is the script you'd like to run and <account> is your TRUBA account.
Three example tensors are given in the directory example_tensors.
You can find more tensors in The Formidable Repository of Open Sparse Tensors and Tools (FROSTT) website. You only need the .tns file for each tensor, not the maps. Simply choose a tensor, download the .tns.gz file, decompress it, and use it with the executable.