Datasets
One of the motivating factors in the design of this benchmark suite was to characterize typical visualization workflows for researchers that use HPC resources and identify bottlenecks. The data-set and workflow scenarios encompass astrophysics, computational fluid dynamics, bioscience, computational fluid dynamics and climate modeling. Some data-sets and visualization workflows provided much greater flexibility and scalability, making them more suitable for the comprehensive and exhaustive tests designed to stress the system components and memory loads. The domain specific benchmarks are launched by a primary python script which will call the appropriate data specific implementation scripts to load the data, apply the requested filters, and visualization algorithms, and render an image. The ability to programmatically vary the geometry and resulting memory footprint was critical for generating useful metrics for system components and comparisons. Similarly, the ability to apply a number of different visualization algorithms or renderers to the same data-set makes it much easier to draw conclusions on what factors effect performance. The results presented here are from the computational dynamics and geoscience datasets and visualization workflows that proved themselves to be the most useful for our purposes. These data-sets provide comprehensive coverage of the visualization operations of interest and can be scaled programmatically to triangle counts from 4 to 300 million. The benchmark suite includes automated graph generation scripts to produce formatted dat files and graphs to highlight the selected variable (e.g. renderer or vis algorithm or system type) as a function of triangle scaling, node scaling, process scaling on one node, and memory utilization. For dynamic data scenarios, these same metrics are used as a function of frame or stage in the workflow.
Datasets for Benchmark Suite
Desricption of Datasets:
DNS
Turbulent Channel Flow: Wall bounded turbulent flow channel flow computed by direct numerical simulation (DNS), meaning, the fluid flow equations are solved without subgrid modeling. Almost a quarter of global energy consumption is expended on transportation through air or water or fluid being transported through ducts, and this energy is dissipated primarily in turbulence. This model of wall-bounded turbulence will help inform the development of new efficient designs of turbulent fluid systems. Each geometry level has a corresponding triangle count in the millions: 69.88, 142.38, 210.33, 276.36, 343.79, 413.39, 485.66, 558.72
Contact: Robert Moser, rmoser@ices.utexas.edu
References: `Petascale direct numerical simulation of turbulent channel flow' Myoungkyu Lee, Nicholas Malaya, Robert D. Moser, proceedings SC13
RM
This simulation is from a very high resolution 3-D simulation of the Richtmyer\--Meshkov instability and turbulent mixing. Richtmyer\--Meshkov instability occurs when two fluids of different density are accelerated. Normally this is by the passage of a shock wave. Small amplitude perturbations cause an instability which initially grow linearly with time. The ability to run at very high resolution is required for the simulation to accurately represent the interaction of the various length scales, and consequently, the reactivity of the intermixing species.\\
Currently all levels have 316.36 million triangles.
Contact: Mark Duchaineau duchaineau@llnl.gov
References: `Very High Resolution Simulation of Compressible Turbulence on the IBM-SP System' A. A. Mirin, R. H. Cohen, B. C. Curtis, W. P. Dannevik et al LLNL tech report UCRL-JC-134237, proceedings SC99
FIU
Groundwater flow through porous materials. Lattice Boltzmann Method flow simulations were computed to simulate the flow of water through porous materials. The fundamental concept is that the conservation of mass and momentum is occurring on a microscopic level where a group of particles located on a lattice interacts by means of two processes known as streaming and collision. The microscopic interaction involves the exchange of momentum as particles collide. During these collisions, mass and momentum are conserved. Visualization of the flow velocities in all 3 directions and density differences were obtained to illustrate the flow paths through the entire domain plotted as tube lines that are colored by the magnitude of the flow velocity.
Contact: Mike Sukop, sukopm@fiu.edu
References: Lattice Boltzmann Modeling and Specialized Laboratory Techniques to Determine the Permeability of Megaporous Karst Rock. S.Garcia. MS thesis, FIU, 2013.
Geo
Morrocan Safi Haute Mer Seismic Survey Horizons. These are horizons are extracted from a volume of seismic data measured from a 1064 sq km seismic survey acquired in the Safi Haute Mer area. The 5 horizons representing features in geologic time, from newest to oldest: Seafloor, Top Albian, Top_MTC and Base_MTC (Mass Transport Deposits), and Basement, were exported from the Geoscience specialty software: Landmarks DecisionSpace as 3D points. Each layer had 4-6 million points and a surface was created using the ball-pivoting surface reconstruction algorithm in the opensource software MeshLab to yield surfaces of 8-10 million triangles per horizon. These surfaces were converted to OBJs before reading the files into Paraview where they were subsequently colored to highlight the geographic features.
Contact: Dallas Dunlop dallas.dunlap@beg.utexas.edu
References: Seismic geomorphology of the Safi Haute Mer exploration block, offshore Morocco Atlantic Margin. D. Dunlop MS thesis. UT Austin, 2013
Molecule
A bacteriophage is a virus that infects a bacterium host and are widely distributed in nature. Sea water is one of the densest natural sources for phages and other viruses. They are seen as a possible therapy against drug resistant strains of many bacteria, and as such the overall shape of the macromolecular assembly is significant. This particular structure (PDB ID 1VRI) is the largest structure in the PDB (150,720 atoms). The ribbon and surface of this structure was computed using VMD and exported as an OBJ for import into Paraview (Paraview does not represent protein secondary structure). This was designed as an example that could be used as a reference in Paraview and will be compared to the VMD Tachyon renderer in phase 2.
Contact: 1VRI.pdb openly available from the Protein Data Bank
References: `Construction and 3-D computer modeling of connector arrays with tetragonal to decagonal transition induced by pRNA of phi29 DNA-packaging motor.' Y.Y.Guo, F.Blocker, F.Xiao, P.Guo