Final thoughts

Summer of HPC 2015 is over now and we have all returned to our homes, eager to work on new projects and using knowledge gained during the past 2 months. There have been some really interesting projects presented during the last Webex conference. In Budapest, our project was mostly about self study in the field of high performance computing. Finally, I had time to learn OpenMP and MPI programming and we had the opportunity to work independently on HPC machines. During our stay, we gained much experience in installation of software packages to HPC machines, the compilation process as well as the installation of libraries, resolving dependencies etc. Finally, we managed to run Quantum Espresso and Siesta and check if everything behaved like it should.

Programs like Quantum Espresso and SIESTA make it possible to perform ab-initio calculations of electronic structure and molecular dynamics. Ab-initio means that structures are calculated using first principles of theoretical physics purely, without any empirical or experimental parameters needed during the calculation. The results of such calculations are the most sophisticated and advanced models of matter currently available, thus enabling us to calculate various physical and chemical properties of materials and molecules, including dynamical properties like vibration and reaction pathways.

A short promotional video, summarizing my work can be seen here:

During the project 4 of the 6 Hungarian supercomputers have been used, namely Budapest2, Szeged, Budapest and Debrecen2. For all of the HPC machines, the SLURM scheduler is used for submitting jobs. Once parameters for calculations are set right, the submitting of jobs is quite easy. For most of the HPC machines we were given reservations so we could experiment straight forward without waiting for our calculations to start. For visualization, three programs have been used extensively: VESTA, XCrysDen and VMD. This are the most common tools in molecular modeling. They are capable of reading many different file formats. However, additional tools have been used for post-processing during visualization of SIESTA output. A lot of specific tools can be found for free on the internet. Since I was supposed to run jobs with different supercomputers and different software I even tested the trial version of Deneb by Atelgraphics. This is a software package designed to manage this kind of work easily. In addition to my local laptop, HPC machines have also been used for visualization, on specialized visualization nodes. To use this resources VirtualGL and VNCViewer have been used.

After getting familiar with SLURM and MPI it was time to learn about the specific SIESTA and Quantum Espresso input format. The directives are quite similar in many of the available ab-initio codes for materials modeling, but often there are details which shouldn’t be left out for the calculations to run as supposed. Sample calculations and examples are probably your best friend when starting to work with a new code of that sort. The compilation of Quantum Espresso and SIESTA turned out to be the most difficult step in the project. For a successful build various libraries have to be linked during the compilation process. This becomes quite confusing very fast, specially if the various configuration options, paths to libraries and hardware specifications are not known or understood exactly.

For a more detailed description of my work at NIIF, you can read my final report. It will be available on the SoHPC website soon.

On our final day, our supervisors at NIIF organised a wonderful Dinner & Cruise event on the Danube. Finally, I would like to share some pictures of our stay in Budapest.