Efficient I/O using Dedicated Cores in Large-Scale HPC Simulations

Large-scale HPC simulations running on cutting-edge machines raise new challenges when it comes to efficiently write, store, analyse and visualize large amounts of data. This PhD research focuses on the Damaris approaches, which leverages dedicated cores on multicore SMP nodes to achieve scalable I/O and data management. We here present past results demonstrating the efficiency of the approach in scaling I/O performance and hiding I/O variability. We then present current work and future directions related to in-situ data analysis and visualization.