In Situ and In Transit Processing and Visualization for Scientific Applications on Hybrid Scalable Architectures:
Simulation has become a very important tool in diverse scientific fields, for instance molecular dynamics, fluid dynamics, biology, materials science, and many more. An important aspect of simulations in the scale at which they run, and the challenges it presents: Modern machines with different architectures (CPUs, GPUs, ARMS, TPUs) are big arrays of interconnected machines, with important concerns about communication, storage and efficiency. However the storage presents an important limitation in its transfer speed.
There have been attempts to minimise the effects of said limitation, in-situ computation and In Transit being one of them. Another important aspect of simulations is visualisation, which allows scientists to analyse the results. Visualisation also enables the scientific community not only to obtain insight and understanding from data but to illustrate and transmit it, and, helps the process of transforming data into information and ultimately, knowledge. The in-situ approach aims to improve efficiency by bypassing storage
entirely, performing data operations in the memory spaces where results are generated. This approach is suitable for many applications, and data visualisation is one of them given its characteristics: it usually produces very large quantities of data and traditionally, it must use storage which generates a substantial overhead in the performance of the complete simulation.
This thematic conference shows our experience in this area, mainly in the projects with the french INRIA-Datamove team and the German HIC Kaisserlauterns research team, applied for scientific applications in hybrid high performance computers and ultrascale systems.
In Situ and In Transit Processing and Visualization for Scientific Applications on Hybrid Scalable Architectures