Abstract:
|
This investigation is held within the GRICCA research group from the UPC (Grup de Reserca e Investigació per la Colaboració Científica Aplicada). The physical location of the computational cluster employed in the project is the GRICCA–lab at the EUETIB (Escola Universitaria d’Enginyeria Tècnica Industrial de Barcelona). The project has been financed by the Ministry of Economy of Spain with two consecutive grants that were used in setting-up the computational facilities.
The computer cluster comprises one master node, which controls job submission, visualization and analysis, and ten computing nodes for the parallelized simulations. Each of these nodes is connected with Infiniband 4x10Gb/s switches (for MPI/HPC) and an Ethernet 1Gb/s connection (for administration and remote log-in). The cluster is equipped with a 500 GB HDD (Hard Disc Drive), 10 x 2.4 GHz-cores and 16 GiB (~109 GB) of RAM (Random-Access Memory). The data storage system is comprised by two 32 TB-discs shared by the master and computing nodes. These technical characteristics ensure optimal parallel calculations among nodes and are ideal for the present investigation.
The above infrastructure allows one to obtain all the results gathered in this study within a reasonable computational time. As an example, one typical simulation submitted to just 1 node would be completed within one month, whereas, by employing the whole parallelized 10 node- system, computational time is reduced to just a few days. The software employed in the atomistic simulations is LAMMPS. The software’s parallelized implementation is extremely efficient, with almost a linear scaling between nodes. |