High-performance astrophysical visualization using Splotch

Abstract

The scientific community is presently witnessing an unprecedented growth in the quality and quantity of data sets coming from simulations and real-world experiments. To access effectively and extract the scientific content of such large-scale data sets (often sizes are measured in hundreds or even millions of Gigabytes) appropriate tools are needed. Visual data exploration and discovery is a robust approach for rapidly and intuitively inspecting large-scale data sets, e.g. for identifying new features and patterns or isolating small regions of interest within which to apply time-consuming algorithms. This paper presents a high performance parallelized implementation of Splotch, our previously developed visual data exploration and discovery algorithm for large-scale astrophysical data sets coming from particle-based simulations. Splotch has been improved in order to exploit modern massively parallel architectures, e.g. multicore CPUs and CUDA-enabled GPUs. We present performance and scalability benchmarks on a number of test cases, demonstrating the ability of our high performance parallelized Splotch to handle efficiently large-scale data sets, such as the outputs of the Millennium II simulation, the largest cosmological simulation ever performed.

Overview on Spectral Line Source Finding and Visualisation

“…They emphasize the need for tools to visualise scientific data in a comprehensive, self-describing, rich and appealing way. SPLOTCH is a powerful package to do this (see also Jin et al 2010).…”

“…SPLOTCH is a powerful and very flexible ray-tracer software tool 4 which supports the visualisation of largescale cosmological simulation data (Dolag et al 2008(Dolag et al , 2011Jin et al 2010). It is publicly available and continues to be enhanced.…”

GPU accelerated particle visualization with Splotch

“…The MPI implementation [27] simply distributes chunks of particles among different processors, each performing a serial computation and producing a partial rendering, the root processor composing final renderings. The OpenMP implementation performs Rasterization exploiting multiple threads each working on different chunks of particles.…”

“…The original Splotch algorithm has been optimised in terms of memory usage and exploitation of standard HPC architectures, e.g. multi-core processors and multi-node supercomputing systems by adopting the MPI paradigm [27] and OpenMP (see Sect. 2).…”

“…As a result we propose a novel approach, based on a concurrent use of the GPU and the CPU. This followed a preliminary investigation, as reported in [27], that had to be redesigned from the ground up to significantly improve performance. A number of further optimised solutions for rendering particles have been implemented, based on a novel data classification and organization strategy.…”