Parallel Flow Visualization

Image

Problem Overview

Particle advection, i.e., displacing particles so that they are tangent to the velocity field, is a foundational element for many visualization algorithms for flow analysis, including streamlines, pathlines, stream surfaces, and Finite-Time Lyapunov Exponents (FTLE) calculation. Particle advection is a particularly difficult form of a non-embarrassingly parallel algorithm, as the work needed to complete the problem is data dependent and thus not known a priori. Further, the workload across particle advection problems can change dramatically. For example, streamline calculation typically involves advecting few particles for long distances, while FTLE calculation typically involves advecting many particles for short distances. Therefore, research on this problem should examine a range of scenarios, considering variation in particle count, distance traveled, and vector field. Finally, visualization and analysis is increasingly being performed in an in situ setting, where visualization and analysis is performed at the same time as the simulation, and using some of its resources. This usage modality increases the need for understanding particle advection over many architectures, and at many concurrencies.

Results

We have published multiple papers on this topic. Highlights include:

  • Exploration-oriented usage from interpolating over Lagrangian tracer particles extracted in situ can be faster, more accurate, and take less storage than the traditional post hoc model where time slices are saved to disk (Agranovsky/LDAV14).
  • Parallelism approaches between parallelize-over-data and parallelize-over-seeds may be better than the commonly used approaches at the two ends of the spectrum (Pugmire/SC09).
  • Hybrid parallelism can benefit flow visualization problems significantly, including speedups of more than 10X on nodes with only four cores (Camp/TVCG11).
  • The ability of an architecture to operate efficiently is dependent on the particle advection workload (Childs/HiPC14, Camp/EGPGV13).
  • SSDs on supercomputers can be used to accelerate some particle advection approaches, specifically those that load data repeatedly (Camp/LDAV11).

People


Sudhanshu Sane
Ph.D. Student

Hank Childs
CDUX Director

External Collaborators

Publications


Improved Post Hoc Flow Analysis Via Lagrangian Representations
Alexy Agranovsky, David Camp, Christoph Garth, Wes Bethel, Kenneth I. Joy, and Hank Childs
IEEE Large Data Analysis and Visualization (LDAV), Paris, France, November 2014
Best Paper

[PDF]     [BIB]

Scalable Computation of Streamlines on Very Large Datasets
David Pugmire, Hank Childs, Christoph Garth, Sean Ahern, and Gunther Weber
ACM/IEEE Conference on High Performance Computing (SC09), Portland, OR, November 2009

[PDF]     [BIB]

Streamline Integration Using MPI-Hybrid Parallelism on a Large Multicore Architecture
David Camp, Christoph Garth, Hank Childs, David Pugmire, and Kenneth I. Joy
IEEE Transactions on Visualization and Computer Graphics (TVCG), November 2011

[PDF]     [BIB]

Particle Advection Performance Over Varied Architectures and Workloads
Hank Childs, Scott Biersdorff, David Poliakoff, David Camp, and Allen D. Malony
IEEE Conference on High Performance Computing, Goa, India, December 2014

[PDF]     [BIB]

GPU Acceleration of Particle Advection Workloads in a Parallel, Distributed Memory Setting
David Camp, Hari Krishnan, David Pugmire, Christoph Garth, Ian Johnson, Wes Bethel, Kenneth I. Joy, and Hank Childs
EuroGraphics Symposium on Parallel Graphics and Visualization (EGPGV), Girona, Spain, May 2013

[PDF]     [BIB]

Evaluating the Benefits of An Extended Memory Hierarchy for Parallel Streamline Algorithms
David Camp, Hank Childs, Amit Chourasia, Christoph Garth, and Kenneth I. Joy
IEEE Large Data Visualization and Analysis (LDAV), Providence, RI, October 2011

[PDF]     [BIB]

Parallel Stream Surface Computation for Large Data Sets
David Camp, Hank Childs, Christoph Garth, David Pugmire, and Kenneth I. Joy
IEEE Large Data Analysis and Visualization (LDAV), Seattle, WA, October 2012

[PDF]     [BIB]

Subsampling-Based Compression and Flow Visualization
Alexy Agranovsky, David Camp, Kenneth I. Joy, and Hank Childs
SPIE Visualization and Data Analysis (VDA), San Francisco, CA, February 2015

[PDF]     [BIB]