Real-time Visualisation of Large Computations
One of the most demanding areas for computations is a simulation of large engineering problems. Distributed computational methods allow using thousands of computational nodes for computation of a single simulation, which means that simulations are becoming more and more exact and useful for estimation of more complex phenomena. One of the disadvantages of this growth is an enormous amount of data which needs to be stored and analysed. Unfortunately, the storage capacity growth does not keep up with the computational power growth of these devices. It means that the ability to generate data has overtaken the ability to store and analyse them. For example, the states of each simulation have to be stored with lower frequency in order to minimize storage capacity requirements.
For simulations, we use ESPRESO (= ExaScale PaRallel FETI SOlver), our unique iterative solver for Exascale systems. For visualisation, we have currently been developing its extension called EXPRESO RTViz, which will use Paraview Catalyst, or possibly OSPRay for real-time visualization of large computations.
While performing coarse time sampling, it is difficult to record a complex behaviour. For solving this problem, we will use Paraview Catalyst, which integrates parallel analysis and visualization directly into simulation codes.
OSPRay (= Open-source Scalable Portable RAY tracing engine) is a freeware display library based on a display method for tracing rays for very demanding interactive visualisation running on Intel architectures (Intel Xeon, Intel Xeon Phi, …).
OSPRay is based on Embree and ISPC (Intel® SPMD Program Compiler) libraries and fully uses new instructional sets such as Intel AVX, SSE, AVX2 nebo Intel® Xeon ™ Phi in order to reach high performance of plotting. Thus, the display speed depends on the CPU and RAM and not on GPU. Employing OSPRay, we will also be able to display models containing a milliard of triangles in real-time (on Salomon on SGI UV 2000 node with 3TB RAM) which is not possible on GPU.