
Anthony Barbu, Tarek Ghaddar, Jonathan Madsen, Carolyn McGraw, Michael Adams, Milan Hanus, Daryl Hawkins, Timmie Smith, Nathan Thomas, Yunhuang Zhang, Marvin Adams, Nancy Amato, Jim Morel, Lawrence Rauchwerger
Project Alumni:
Ping An, Teresa Bailey, Jae Chang, Tao Huang, Josh Jarrell, Alin Jula, Alex Maslowski, Mark Mathis, Olga Pearce, Silvius Rus, Armando Solar, Lidia Smith, Gabriel Tanase, Mauro Bianco
The overall goal of the project is the development of strategies that will produce the desired discreteordinates transport solutions in the lowest possible wallclock time on computational platforms of interest to the DOE's Accelerated Strategic Computing Initiative (ASCI). The project began in 1998 with the best known sequential algorithms for solving the discreteordinates transport problem as starting points for devising algorithms and coding strategies to yield the best possible performance on modern parallel cachebased systems. To date the project has produced a transport code (PDT) that is capable of solving steadystate discreteordinates problems using regular hexahedral grids and multiple transport algorithms and acceleration methods. Work to allow the code to handle arbitrary grids and to solve timedependent transport problems is in progress.
Problem Statement
The deterministic transport problem solved by PDT can be described briefly as:
Given:
Compute:
The flow of particles at a subsequent time at every point in the domain.
PDT is written in C++ using STAPL. The code represents the problem as a set of objects described below. The main function of the application uses the constructors and initialization methods of the objects to setup the problem and then calls the solve method of the problem class.
Generic Problem
We represent a problem as a system containing:
Generic Grid and ElementMap
The Grid object represents the topology of the given spatial discretization. It is made of cells.
In order to solve the problem we introduced a parallel topology based on computational elements rather than spatial ones. That is the ElementMap. It is made of elements.
Parallel Solver
We have already introduced the concept of ElementMap, which was created to ease the abstraction of solving the problem. There are other constructs that ease the development of solvers for parallel machines:
Generic Chunk
The Chunk is composed of:
It is the computation unit for sweeping algorithms. A chunk is the atomic unit of work in the sweep. For distributed memory systems the chunks are also communications atoms since messages carrying information to cells on different processors are buffered until all cells in the set are processed..
Partitioner and Scheduler
The Partitioner reads the problem size from the input file and determines the assignment of cells to each thread of execution. The Scheduler accepts a set of dependence graphs on Cells as input and aggregates the cells into cell sets. The Scheduler also produces dependence graphs based on cell sets and angle sets.
Generic Executor
The executor is implemented as a pAlgorithm in
STAPL named p_for_all.
The algorithm takes as input a set of dependency graphs on Chunks (which are
represented using the STAPL pRange) and a generic function to be executed on
every Chunk. It manages parallel execution by determining the next
available Chunk to process on each thread based on the dependence graphs
given. On distributed memory systems it manages the communication too
by forcing any buffered messages to be sent after each call of the work
function given as input.
Researchers at Lawrence Livermore National Laboratory, Los Alamos National Laboratory, and Sandia National Laboratory are working on deterministic transport calculations. In order to avoid duplication of effort, foster collaboration and crossfertilization, and expose students and researchers to one another, there has been a series of workshops held at Texas A&M University. At each of the workshops researchers from the labs and students working on the project would present their work on their respective projects.
Labfest 3, May 2001
Labfest 4, November 2003
Labfest 5, April 2004
Labfest 6, May 2005
Papers
W Hawkins, Timmie Smith, Michael Adams, Lawrence Rauchwerger, Nancy Amato, Marvin Adams, "Efficient Massively Parallel Transport Sweeps," Trans. Amer. Nucl. Soc., 107(1):477481, Nov 2012.
Journal(pdf)
Roger Pearce, Bryan Boyd, Xinyu Tang, Darla Haigler, Akhil Patel, Nancy M. Amato, "Supporting Path Planning Queries Incorporating Multiple Modes of Transportation using Layered Roadmaps," Technical Report, TR06014, Parasol Laboratory, Department of Computer Science, Texas A&M University, Oct 2006.
Mark M. Mathis, Darren J. Kerbyson, Adolfy Hoisie, "A Performance Model of nonDeterministic Particle Transport on LargeScale Systems," Future Generation Computer Systems, 22(3):324335, Feb 2006. Also, In Proc. Int. Conf. on
Computational Science (ICCS), Melbourne, Australia, Jun 2003.
Journal(pdf, abstract) Proceedings(ps, pdf, ppt, abstract)
Mark M. Mathis, Darren J. Kerbyson, "A General Performance Model of Structured and Unstructured Mesh Particle Transport Computations," Journal of Supercomputing, 34(2):181  199, Nov 2005.
Journal(pdf, abstract)
Mark M. Mathis, Darren J. Kerbyson, "Performance Modeling of Unstructured Mesh Particle Transport Computations," In Proc. Int. Par. and Dist.
Proc. Symp. (IPDPS), Santa Fe, NM, Apr 2004.
Proceedings(ps, pdf, ppt, abstract)
William McLendon, Bruce Hendrickson, Steve Plimpton, Lawrence Rauchwerger, "Finding Strongly Connected Components in Parallel in Particle Transport Sweeps," In Proc. ACM Symp. Par. Alg.
Arch. (SPAA), pp. 328329, Crete, Greece, Jul 2001.
Proceedings(ps, pdf, abstract)
Mark M. Mathis, "A General Performance Model for Parallel Sweeps on Orthogonal Grids for Particle Transport Calculations," Masters Thesis, Department of Computer Science and Engineering, Texas A&M University, Dec 2000.
Masters Thesis(ps, pdf, abstract)
Mark M. Mathis, Nancy M. Amato, Marvin Adams, "A General Performance Model for Parallel Sweeps on Orthogonal Grids for Particle Transport Calculations," In Proc. ACM Int. Conf.
Supercomputing (ICS), pp. 255263, Santa Fe, NM, May 2000. Also, Technical Report, TR00004, Parasol Laboratory, Department of Computer Science, Texas A&M University, Dec 1999.
Proceedings(ps, pdf, abstract) Technical Report(ps, pdf, abstract)
Nancy M. Amato, Ping An, "Task Scheduling and Parallel MeshSweeps in Transport Computations," Technical Report, TR00009, Department of Computer Science and Engineering, Texas A&M University, Jan 2000.
Technical Report(ps, pdf)
Parasol Home  Research  People  General info  Seminars  Resources Parasol Lab, 301 Harvey R. Bright Bldg, 3112 TAMU, College Station, TX 778433112 parasoladmin@cse.tamu.edu Phone 979.458.0722 Fax 979.458.0718 Department of Computer Science and Engineering  Dwight Look College of Engineering  Texas A&M University Privacy statement: Computer Science and Engineering Engineering TAMU Web Accessibility Policy and Law  Web Accessibility and Usability Standards  Contact Webmaster 