Focus now on the correct implementation of the matrix multiplication you find in https://github.com/SimoneReale/SE4HPC_project_part2. This is a parallel implementation that uses MPI and reads the matrices to be multiplied from two files, matrixA.txt and matrixB.txt. In these files the first row contains the matrix dimensions (number of rows and columns), while the other rows contain the matrix itself.
Your task is to perform the following steps:
Preparation: Use the template available here https://github.com/SimoneReale/SE4HPC_project_part2 to create your own github repository. Add to this repository the tests you have created in Step1.
Automating the build, test and release processes: Create a CI/CD pipeline that, when someone pushes files in the repo, executes the building and testing process.
Containerizing the application: Go through the following steps:
-
Define a Singularity container descriptor for the matrix multiplication program and push it in your repo.
-
Extend the created action to create a container image from your description.
Executing on the cluster: Go through the following steps:
-
Create a job.sh file to run your containerized application. Make sure that the standard output and error are mapped to txt files.
-
Transfer on Galileo100 your job script and the container.
-
Submit your job to the cluster and check whether it works correctly.
-
Push on your github repository your job.sh file and the files obtained from the execution of the matrix multiplication.
Extend the action you have created at step 3 to automate completely the process from a push on the repository to the execution of the containerized software on SLURM. To do so, you will have to move your container from the runner to the cluster. You can either use the scp command or you can publish your image on the Singularity registry and then pull it from the cluster. Don't forget to handle your secrets properly! You do not want to leave passwords and authentication tokens visible to everybody, so you will use the secrets mechanism.