@mastersthesis{christoph_honal_thesis.pdf, author = {Christoph Honal}, title = {Efficient I/O Strategies for Checkpointing and Visualization in sam(oa)²}, year = {2019}, school = {Technical University Munich}, month = {Mar}, language = {en}, abstract = {The implementation of large-scale finite-element simulations on distributed systemsrequires specific memory management and disk I/O handling. This is due to theimmense spatial and temporal resolutions and the grade of heterogeneous paralleliza-tion, i.e. hybrid MPI/OpenMP. In this thesis, a new approach for storing, visualizingand restoring (referred to as checkpointing) simulation data in the sam(oa)2project ispresented, introducing a parallel storage system based on the HDF5 and XDMF fileformats. The presented implementation measurably reduces disk space usage and filesystem strain, while improving run-time performance compared to previous outputformats.}, }