The Zoltan Team

Past Zoltan Contributors

• Zoltan v3.3 was released in July 2010 as part of the Trilinos project. You can download as a stand-alone package or in the Trilinos distribution.
  
  
• "Getting Started with Zoltan: A Short Tutorial."
  A ten-page introduction to using Zoltan is now available. Download it here:
  PDF         Citation (BIBTEX)         Viewgraphs from the tutorial (PDF)
  
  
• The Zoltan Tutorial presented at SciDAC2007 is now on-line. Download it here:
  • Zoltan Tutorial Slides (14 Mbytes; PDF)
  • Zoltan Tutorial Handouts (Four slides per page; 8.8 Mbytes; PDF)

• The Zoltan Frequently Asked Question (FAQ) list is now on-line.
• The Zoltan team received the "Best Algorithms Paper Award" at the 2007 International Parallel and Distributed Processing Symposium (IPDPS07) for their paper "Hypergraph-based Dynamic Load Balancing for Adaptive Scientific Computations."
  Abstract (HTML)      Paper (PDF)      Citation (BIBTEX)

• Zoltan is part of two funded DOE SciDAC projects:
  • CSCAPES: Combinatorial Scientific Computing and Petascale Simulations
    Contact: Erik Boman
  • ITAPS: Interoperable Technologies for Advanced Petascale Simulations
    Contact: Karen Devine
    

New Features in Zoltan v3.3, released July 2010: Release Notes       |       Backward Compatibility

• New local ordering method based on space-filling curves to improve memory and cache locality within a processor.
• Ability to call graph partitioning algorithms using hypergraph callback functions; this capability is useful applications with, say, block-structured matrix distributions (e.g., SuperLU), where all information about a matrix row or column is not available on a single processor.
• Improved execution time of parallel hypergraph partitioning.

New Features in Zoltan v3.2, released September 2009: Release Notes       |       Backward Compatibility

• New interface to Scotch and PT-Scotch parallel graph partitioning algorithms.
• Simplified interface to graph ordering and coloring algorithms
• Automated symmetrization of graphs for graph partitioning, coloring and ordering. (See parameters GRAPH_SYMMETRIZE and GRAPH_SYM_WEIGHT in the Scotch and ParMETIS graph packages.)
• Improved function Zoltan_LB_Eval returns more information about a decomposition to users.
• Improved examples showing Zoltan usage in C and C++ are included in zoltan/example.
• Improved support for builds under autotools, including builds of Zoltan's F90 interface.
• New support for CMake builds and testing through Trilinos; builds of Zoltan's F90 interface are included.
• Improved integration into Isorropia partitioners for Trilinos' Epetra classes.

New Features in Zoltan v3.1, released September 2008:

• Graph/Matrix ordering interface to PT-Scotch, a high-quality graph ordering and partitioning library from LaBRI/INRIA-Bordeaux.
• New matrix ordering interface that returns ordering information such as permutations and separators to the application.
• New hypergraph partitioning options for inexpensively refining partitions.
• Robustness improvements to Zoltan's parallel hypergraph partitioner and repartitioner.
• New Autotools build environment.
• Tight integration with the Trilinos project, including Trilinos package Isorropia, a matrix-based interface to Zoltan.
• Serial, non-MPI builds are now available.

• Parallel Hypergraph Repartitioning and Dynamic Load Balancing to partition data while accounting for current processor assignments, thus reducing the amount of data migration needed; see the Zoltan User's Guide for details.
• Hypergraph refinement to quickly improve the quality of an existing partition.
• Hypergraph Partitioning and Repartitioning with Fixed Vertices allows users to specify the partition assignment of objects.
• Native graph partitioning using Zoltan's hypergraph partitioner.
• Hierarchical partitioning for hybrid partitions on heterogeneous computers.
• Improved part remapping to reduce data migration costs.
• Simple block and random partitioners for demonstration and testing.

New Features in Zoltan 2:

• Parallel Hypergraph Partitioning
• Distance-1 and Distance-2 graph coloring
• Multicriteria Geometric Partitioning
• C++ Interface

• Release Notes
• Backward Compatibility
• Download Now!

• a suite of dynamic load-balancing and parallel partitioning tools that distribute data over sets of processors,
• data migration tools that simplify movement of data to new parts,
• parallel graph/matrix ordering algorithms for fill-reducing matrix ordering, including interfaces to PT-Scotch (LaBRI/INRIA-Bordeaux) and ParMETIS (U. Minnesota),
• parallel graph coloring including distance-1 and distance-2 coloring,
• distributed data directories that efficiently locate off-processor data,
• an unstructured communication package that greatly simplifies interprocessor communication, and
• a dynamic memory debugging package for memory-debugging on state-of-the-art parallel computers.

Why Zoltan is needed:

• In some applications, work loads and/or geometric locality change as computations proceed;
  Zoltan provides dynamic redistribution of data to maintain high performance. 
• Adaptive finite element methods
• Particle methods
• Contact detection algorithms
• Multiphysics simulations
• Adaptive physics models
• Processors need to track off-processor data's locations in dynamic environments; Zoltan's distributed data directories allow applications to efficiently query and update off-processor data locations.
• Unstructured and adaptive applications require complicated interprocessor communication; Zoltan's unstructured communication package manages sends and receives for application developers.
• State-of-the-art parallel computers often do not have sophisticated debugging tools available; Zoltan's dynamic memory management package simplifies location of memory leaks and other memory errors.

• Zoltan's object-oriented interface is clean and easy to use in applications.
• A suite of load-balancing algorithms is included in Zoltan;
  try them all to see what works best in your application.
• Interfaces to the popular ParMETIS and Jostle packages are included.
• Application development time can be reduced by using Zoltan's parallel utilities for unstructured communication, dynamic memory management, data migration, and distributed data directories.

• New algorithms can be added to Zoltan in a straightforward manner.
• Zoltan's object-based interface provides a simple, general way to get needed application data to the algorithms.
• New algorithms can be compared to high-quality implementations of existing algorithms.
• Zoltan's data-structure neutral design allows new algorithms to be used easily in many different applications.

• Overview
• Design Considerations
• Dynamic Load-Balancing and Data Migration Tools
• Communication Tools
• Memory Management Tools

• User's Guide (HTML, PDF)
• Developer's Guide (HTML, PDF)

• Publications
• Papers and Presentations
• Serial hypergraph partitioners
• PaToH (Ohio State University)
• Mondriaan (Utrecht University)
• hMetis (University of Minnesota)
• Parallel hypergraph partitioners
• Parkway (Imperial College, London)
• Parallel and dynamic partitioners
• ParMETIS (University of Minnesota)
• Jostle (University of Greenwich)
• DRAMA (ESPRIT IV)
• Serial graph partitioners
• Chaco (Sandia National Laboratories)
• METIS (University of Minnesota)
• Party (University of Paderborn)
• SCOTCH (Laboratoire Bordelais de Recherche en Informatique, Université Bordeaux I)
• Sandia National Laboratories
• Sandia's Computation, Computers and Math Center
• Sandia's Computer Science Research Institute