Tiger Team Projects
The following enumeration provides an overview of collaborations between members of the bwHPC-S5 team and scientists, i.e. tiger teams. To apply for support by a tiger team, click
Optimization of Molcas scratch data management on bwForCluster JUSTUS
Group of Molecular Quantum Devices of Institute of Nanotechnology on KIT uses Molcas to determine magnetic properties of functionalized organometallic complexes by modern methods of quantum chemistry. Correct explanation of structure/functional relationship of various organometallic compounds typically includes rationalized interpretations of experimental measurements as well as theoretical elucidations of the specific magnetic phenomena at the atomic scale level. This usually requires to adopt ab-initio methods in order to determine magnetic moments with the corresponding energy-level diagrams and by consideration of contributions from various relaxation mechanisms. Such processes are time and computationally intensive tasks since they include structure optimizations with subsequent post-processing of the electronic structures for relatively large molecules. Due to such complexity an accomplishment of a typical computational task within one job execution on HPC system with limited walltime is not feasible and an optimization of the restart procedure might thus noticeably reduce time to generate scientific outcomes. The main goal of the project was to optimize restart procedure of the Molcas simulations of the large-scale molecular systems on bwForCluster JUSTUS. Adopted strategy defined two particular objectives: (i) to implement suitable method for an effective identification of the latest possible checkpoints from which a subsequent restart can be proceeded without a need of redundant recalculation of already computed data and (ii) to optimize transfer management of scratch data to proceed a subsequent Molcas job. In order to achieve this we have designed an user-specific sub-shell script which upon its utilization in the submission moab-script performs periodical monitoring of the scratch data during the run of the job, estimate the ideal checkpoint and realize the final redistribution of the scratch data to be ready for the consecutive submission. Practical applications has shown an increase of the total job throughput around 40% with respect to a situation before the project started. Finally we have made a plan for a further continuation of the project in 2019 where we want to test and implement memory dumping and restoring technology for Molcas jobs.
Members of the Tiger-Team: Institute of Nanotechnology - Molecular Quantum Devices, Karlsruhe Institute of Technology; Competence center for computational chemistry and quantum sciences, University of Ulm