David DeMarle


M.S. in Computer Science, 2009
Master's Thesis: "Distributed Interactive Ray Tracing for Large Volume Visualization"
Advisor: Steve Parker

David DeMarle is the focus of this quarter's SCI Alumni Highlight. David received his M.S. in Computer Science from the University of Utah in 2003. In 2004, Dave was awarded an internship at the Los Alamos National Laboratory. His Master's thesis focused on the design and use of a distributed shared memory (DSM) system to render data sets that require the aggregate physical memory space of a cluster. As a Teaching Assistant in Ross Whitaker's Scientific Visualization course and while doing his research into parallel rendering with Chuck Hansen and Steve Parker, David honed the skills needed to perform his current role.

David joined Kitware as a Research and Development Engineer in June of 2005. He primarily contributes to expand the capabilities of ParaView and VTK, open-source scientific visualization tools that are used by thousands of researchers around the world and are integrated in many scientific applications. David's secondary responsibilities at Kitware include support for its large user and developer communities. In this role, he answers a wide variety questions on the VTK and ParaView mailing lists and teaches courses about Kitware's tools at conferences around the world.

In ParaView and VTK, David has made significant, formative contributions to:
  • selection - the ability to identify and extract salient features from complex datasets either with (by data queries) or without (by mouse interaction) a priori knowledge of what is interesting),
  • scriptability - the ability to dynamically create, at run time, general purpose data processing operations that are run in parallel on the world's largest supercomputers by leveraging the python programming language inside of parallel VTK filters,
  • streaming - the ability to process data in pieces, and at varying resolution, so as to quickly obtain high level insights and then to progress to full resolution accuracy while never overwhelming the memory limits of the computing resources at hand,
  • advanced rendering - interfacing VTK to the University of Utah's Manta ray tracer, to achieve high quality yet interactive renderings of large (billions of unstructured cells) parallel data sets,
  • and GPGPU. In this last development, he is currently collaborating with Los Alamos National Lab on the development of an infrastructure to leverage modern, many-core architectures within VTK pipelines.

For more information on David's publication, click here.

Richt-MeshCFD joined lucyin
CFD data set rendered interactively with dirt using the aggregate memory space of a 32 node cluster in the lab. The left is an unmodified view of the data, the top right shows the screen space work decomposition, the lower right shows the data space work decomposition. Optimization of the memory layout of large meshes. The first image shows the memory layout in the input mesh. The second shows the memory layout after preprocessing code.
MantaView window running inside ParaView.