ZoltanHandsOnTutorial

Zoltan Hands-on Tutorial

First things first:

• The Zoltan library is a collection of data management services for parallel, unstructured, adaptive, and dynamic applications. It simplifies the load-balancing, data movement, unstructured communication, and memory usage difficulties that arise in dynamic applications such as adaptive finite-element methods, particle methods, and crash simulations.
• Zoltan can be downloaded as a standalone library from the http://www.cs.sandia.gov/~web1400/1400_download.html. Zoltan is also part of the Trilinos project. For information on the Trilinos project visit the official Trilinos website.
• For future reference, a lot of information can be found on the Zoltan page.

Let's dive in, you have two options:

• WebTrilinos

  • WebTrilinos is a scientific portal, a web-based environment to use several Trilinos packages through the web.
  • Note: Only packages that Trilinos was configured with can be used through the WebTrilinos interfaces.
  • Installed at St. John's University, MN. The following versions of the C++ interface are available (the examples below use the C++ interface):
    • Trilinos development branch (10.7), MPI-enabled
    • Trilinos 10.6, MPI-enabled
    • Trilinos 10.4, serial

• Trilinos 10.7 (dev), CUDA-enabled

  • In addition to the C++ interfaces, a Matrix Portal interface and a Python interface are also available.
  • Access to this site is password protected. Login information will be given during live tutorials as needed.
  • Reminder: Use Ctrl+A to highlight all the example code, Ctrl+C to copy it, and Ctrl+V to paste it in the WebTrilinos window.
  • Download and build the examples on your machine
    1. Download Trilinos
    2. Download and install CMake (if not already done).
    • The latest CMake release can be downloaded from here.
    1. Download and install Clapack (if you don't have LAPACK and BLAS already).
    • Get Clapack from here.
    • Use CMake to build Clapack.
    1. Build and install Trilinos.
    • An example script is here.
    • You can find other scripts in the Trilinos distribution Trilinos/sample_scripts directory.
    • You can also use the CMake gui, or the text-gui ccmake.
    1. Get Makefile from here.
    • Learn about the Makefile and the Makefile.export system here.
    • You can find the Makefile.export.package_name files once you have built and installed Trilinos.
    • They will be in the include directory of your installation directory.
    • Example: TrilinosInstall/include/Makefile.export.Epetra.
    1. Customize the Makefile to your situation (the example Makefile uses only the Epetra package, but Makefile.export.package_name files are available for all packages).
    2. Add Example code into a file called main.cpp and build away!

Now for some examples:

1. Load balancing using Zoltan * Example that uses Recursive Coordinate Bisection in Zoltan to partition a mesh input * Example that uses Graph Partitioning in Zoltan to partition a graph input * Example that uses Hypergraph Partitooning in Zoltan to partition a hypergraph input * Example that uses Graph Repartioning in Zoltan to repartition a graph input * Example that uses Graph Partitioning in Zoltan to partition a graph input and migrate the data

2. Matrix Partitioning using Isorropia * Example that partitions an Epetra matrix in Trilinos and redistributes the matrix

Learning more:

• Zoltan website.
• Zoltan User Guide.
• Zoltan Related Papers and Presentations
• Isorropia website