Scientific Visualization and HPC Applications: Minisymposium Abstract

Abstract

High Performance Computing (HPC) produces enormous amounts of data. This simple truth has been the perennial bane of the HPC user and there is no sign of the problem going away. The results of the computational process are often large data sets in the form of molecular structures and property fields, fluid density and velocity fields, particle positions and momenta or any of a diverse host of other types all sharing the single property that they are large and so difficult to interpret. In the case of many computational methods it is, in addition, often useful to retain state information, perhaps as large as the final output, from each step of the computational process for a post-mortem analysis of the optimization or for computational steering. The size of the data produced often scales with the problem size, the problem size typically increases with the available computational power and so the ever-growing improvement in computational power is likely to continue to make this problem more difficult as time progresses.