1. Vijay Mahadevan
  2. MOAB

Overview

HTTPS SSH

MOAB: Mesh-Oriented datABase

MOAB is a component for representing and evaluating mesh data. MOAB can store structured and unstructured mesh, consisting of elements in the finite element "zoo". The functional interface to MOAB is simple yet powerful, allowing the representation of many types of metadata commonly found on the mesh. MOAB is optimized for efficiency in space and time, based on access to mesh in chunks rather than through individual entities, while also versatile enough to support individual entity access. MOAB can be used in several ways:

  • As the underlying mesh data representation for applications
  • Several computational solvers in various scientific domains (nuclear engineering, nonlinear thermo-mechanics, CFD, etc)
  • VERDE mesh verification code
  • Mesh quality computation
  • As a mesh input mechanism (using mesh readers included with MOAB),
  • As as a translator between mesh formats (using readers and writers included with MOAB).

MOAB was developed originally as part of the CUBIT project at Sandia National Laboratories, and has been partially funded by the DOE SciDAC program (TSTT, ITAPS, FASTMath) and DOE-NE (NEAMS program).

Dependencies

  • MPI: MOAB supports usage of MPICH and OpenMPI libraries configured externally in order to enable scalable mesh manipulation algorithms.
  • HDF5: In order to manage the data dependencies and to natively support parallel I/O, MOAB uses a custom file format that can represent the entire MOAB data model in a native HDF5-based file format. Support for this file format requires version 5 of the HDF library, which can be obtained at HDF5.
  • NetCDF: MOAB library optionally depends on the NetCDF libraries (C and C++) to compile the ExodusII reader/writer. To get netcdf, go to NetCDF.
  • Metis: MOAB can use the Metis library for partitioning mesh files in serial
  • Zoltan: Support for online partitioning through Zoltan (and its dependencies on Scotch, Parmetis etc) can be utilized through the partitioner tool

Configuration and Build

  • Unpack the source code in a local directory.
  • Run autoreconf -fi to generate the configure script
  • Run the configure --help script in the top source directory to see a list of available configure options.
    • Use --prefix=INSTALL_DIR to specify installation directory
    • Override default compilers with environment or user options: CC, CXX, FC, F77
    • If you have MPI installed, use --with-mpi=$MPI_DIR
    • If you have HDF5 and NetCDF installed, use --with-hdf5=$HDF5_DIR and --with-netcdf=$NETCDF_DIR to specify external dependencies.
    • Auto-Download options: MOAB now supports automatic dependency download and configuration that has been tested on various platforms and architectures.
      • HDF5: Use --download-hdf5 OR --download-hdf5=TARBALL_PATH
      • NetCDF: Use --download-netcdf OR --download-netcdf=TARBALL_PATH
      • Metis: Use --download-metis OR --download-metis=TARBALL_PATH
  • Now run the configure script with desired configuration options either in-source or out-of-source (build) directory.
  • In the build directory, run the following:
    • Compile MOAB library and supported tools: make -j4.
    • Verify configuration and build setup: make check.
  • To install the compiled libraries, headers and tools, run make install.
  • You can now use the makefile generated under the build/examples folder and modify it to compile user code dependent on MOAB libraries

Continuous Integration

There are several hooks to online continuous integration systems, nightly and commit-based Buildbot/Bitbucket builds that are constantly run during a development day to check the integrity and robustness of the MOAB library.

Current overall build status

  • Buildbot: Buildbot Status

  • Bitbucket Builds (beta): Bitbucket Build Status

  • CodeShip: Codeship Build Status

  • Drone: Drone.io Build Status

  • Code Coverage:

    • Coverity: Coverity Scan Build Status
    • CodeCov: codecov

Bugs, Correspondence, Contributing

MOAB is LGPL code, and we encourage users to submit bug reports (and, if desired, fixes) to moab-dev@mcs.anl.gov. Users are encouraged to check SIGMA-MOAB documentation pages often for news and updates. Please submit pull requests (PR) with a Bitbucket fork or send us patches that you would like merged upstream.


Detailed documentation


Where's the configure script?

After installing the GNU autotools suite, execute the following command to generate the configure script (and other necessary generated files):

    autoreconf -fi

If for some reason, the autoreconf command is not available, the following sequence of commands should have the same result:

    autoheader
    aclocal -I m4
    libtoolize -f
    autoconf 
    automake -a
  1. Why aren't the configure script and other generated files in the CVS repository?

    Because they are generated files. Why save a version history for them? Further, some of the above commands get re-run automatically when Makefile.am's or other files are changed. This could lead to compatibility problems if some of the generated files in the Git repository are from a different version of the GNU autotools.

  2. Aren't we requiring users to have GNU autotools installed in order to configure MOAB?

    No. Developers (or anyone else using source directly from the Git repository) must have the autotools installed. When creating a tarball for distribution of MOAB, the commands below should be run. The resulting tarball will contain all necessary generated files, including the configure script.

  3. What needs to be done to prepare MOAB for distribution as a tarball?

    • Check out a clean copy of MOAB.
    • Execute the following commands in the top-most directory:
      • autoreconf -fi
      • ./configure
    • To create a distributable tarball from a working source directory, do make dist

MOAB iMesh Interface Implementation, iMesh v1.2

A. The list of non-compliant iMesh functionality are enumerated below.

A. 1. Iterators for list-type entity sets:

The iMesh 1.2 specification requires that iterators over list-type entity sets be updated in response to membership changes in the set. Specifically, if entities are added to or removed from the set, the spec requires that added entities be iterated without needing to reset the iterator, and that removed entities not be iterated. MOAB will not support this capability in the iMesh 1.2 release. Future support will depend on whether this can be implemented efficiently, without degrading performance for static mesh applications.

A. 2. No support for septahedron entities:

MOAB does not support septahedron entities at this time (though such entities could be represented as general polyhedra).

B. MOAB capabilities not accessible through iMesh:

B.1. Dense tags: MOAB supports two kinds of tag storage: dense tags, where tag values are stored in sequence for sequences of contiguous entity handles; and sparse tags, which are stored in (entity handle, tag value) tuples. iMesh does not support creation of a tag with a default value, nor does it have a mechanism for passing general options to the tag creation function. Therefore, MOAB's iMesh implementation creates sparse tags by default. Alternatives for specifying the tag creation type will be explored for future iMesh releases.

B.2. Variable-length tags: MOAB supports a variable-length tag, where a tag value can have a different length for each entity to which it is assigned. This functionality is not supported in iMesh.

B.3. Direct access to tag data (tag_iterate functionality): MOAB 4.x introduced the ability for applications to get direct access to tag storage for dense-type tags (see the tag_iterate function in src/moab/Interface.hpp). This functionality is not supported in iMesh.

B.4. Corner vs. all vertices in connectivity list: MOAB represents so-called "higher-order entities", e.g. quadratic tetrahedra, by allowing the connectivity list to be an application-specified size. The connectivity array returned by MOAB's iMesh implementation will always be the total number of vertices, including any high-order vertices. MOAB's interface allows applications to specify whether all or just corner vertices are requested.

B.5. Retrieval of entity, set handles in order from list-type sets: MOAB uses the same handle type for both entities and entity sets. The order of addition of entities and entity sets to list-type sets is therefore preserved, since these handles are stored in the same list. Since iMesh uses a different handle type for entities than for sets, and returns those handles in different API functions (iMesh_getEntities versus iMesh_getEntSets), the original order of addition in the case where entities and entity sets are interspersed cannot be recovered.

B.6. No support for knife-type elements: MOAB supports a seven-sided element referred to as a "Knife" element; this element results from collapsing a quadrilateral bounding a hexahedron by merging opposite nodes. This element type is not supported in iMesh.


Supported File Formats

Some of the file formats listed below may not be supported by a particular build of MOAB depending on the availability of external libraries. An up-to-date list of file formats supported in a particular build of MOAB can be obtained programatically using the MBReaderWriterSet API or as a simple list using the '-l' option of the mbconvert utility.

Format               Name     Read    Write   File name description
------------------  ------  -------- -------  ----------------
MOAB native (HDF5)  MOAB      yes      yes     h5m mhdf
Exodus II           EXODUS    yes      yes     exo exoII exo2 g gen
Climate NC          NC        yes      yes     nc
IDEAS Format        UNV       yes      no      unv
MCNP5 Format        MESHTAL   yes      no      meshtal
NASTRAN format      NAS       yes      no      nas bdf
Abaqus mesh format  ABAQUS    yes      no      abq
Kitware VTK         VTK       yes      yes     vtk
RPI SMS             SMS       yes      no      sms
Cubit               CUBIT     yes      no      cub
QSlim format        SMF       yes      yes     smf
SLAC                SLAC      no       yes     slac
GMV                 GMV       no       yes     gmv
Ansys               ANSYS     no       yes     ans
Gmsh                GMSH      yes      yes     msh gmsh
Stereo Lithography  STL       yes      yes     stl
TetGen mesh files   TETGEN    yes      no      node ele face edge
Template input      TEMPLATE  yes      yes     

Any of the values from the Name column may be passed as an option to the file I/O methods to request a particular file. If no file format is specified, the default is to choose the write format using the file extension and to try all file readers until one succeeds.


File IO Options

An options list as passed to MOAB file IO routines is a single C-style string containing the concatenation of a list of string options, where individual options are separated by a designated separator character. The default separator character is a semicolon (;). To specify an alternate separator character, begin the options string with a semicolon followed by the desired separator. Options are not case sensitive.


Common Options

PRECISION=<N>: Specify the precision to use when writing float and double values (such as node coordinates) to text-based file formats.

CREATE: Do not overwrite existing file.

FACET_DISTANCE_TOLERANCE=<D>: Max. distance deviation between facet and geometry, default:0.001.

FACET_NORMAL_TOLERANCE=<N>: Max. normal angle deviation (degrees) between facet and geometry, default:5.

MAX_FACET_EDGE_LENGTH=<D>: Max. facet edge length, default:0.

CGM_ATTRIBS={yes|no}: Actuation of all CGM attributes, default:no.

DEBUG_IO=n: Set threashold for debug messages from low-level (format-specific) reader/writer. Default is 0 (none).


Parallel IO Options

MOAB must be built with parallel support enabled before these options can be used.

Parallel Read Options:

PARALLEL={NONE|BCAST|BCAST_DELETE|READ_DELETE|READ_PART}: Set parallel read mode.

Options are: - NONE - force serial read/write on each processor (default) - BCAST - read on one processor and broadcast a copy to all others - BCAST_DELETE - read file on one processor, broadcasting partitioned data to other processors. - READ_DELETE - read file on all processors and partition by deleting mesh from non-local partitions. - READ_PART - read only relevant subset of file on each processor, utilizing format-specific parallel I/O if available (this option is not supported for all file formats.) - WRITE_PART - write only relevant subset of file on each processor, creating a single file utilizing format-specific parallel I/O if available (this option is not supported for all file formats.) - FORMAT - deprecated (use WRITE_PART)

PARALLEL_RESOLVE_SHARED_ENTS: Resolve which entities are shared between which processes, such that propogation of data across processes can be done. This should probably be the default behavior, as you almost certainly want this unless, perhaps, PARALLEL=BCAST.

PARTITION OR PARTITION=<tag_name>: Specify that mesh should be partitioned using partition IDs stored in a tag. If the tag name is not specified, the default ("PARTITION") is used.

PARTITION_VAL=<int_list>: If a tag name is specified to the 'PARTITION' option, then treat as partitions only those sets for which the tag value is a single integer for which the value is one of the integers in the specified list.

PARTITION_DISTRIBUTE: Deprecated. Implied by "PARTITION" option.

PARTITION_BY_RANK: Assume 1-1 mapping between MPI rank and part ID. Assigning parts to processors for which rank == part id.

MPI_IO_RANK=<RANK>: For IO modes in which a single processor handles all disk access, the MPI rank of the processor to use. Default is 0.

PARALLEL_COMM=<id>: Specify moab::ParallelComm::id() == <id> to use as communicator.


Format-specific options

Stereo Lithography (STL) files

BINARY|ASCII: Write binary or text STL file. Default is text.

BIG_ENDIAN|LITTLE_ENDIAN: Force byte ordering of binary data. Default is BIG_ENDIAN for writing and autodetection for reading (BIG_ENDIAN if autodetect fails).

MOAB Native (HDF5-based MHDF) format

ELEMENTS={EXPLICIT|NODES|SIDES}: If reading only part of a file, specify which elements to read in addition to those contained in the specified set. The options are:

  • EXPLICIT - read only explicitly designated elements
  • NODES - read any element for which all the nodes are being read.
  • SIDES - read explicilty specified elements and any elements that are sides of those elements.

Default is SIDES unless only nodes are explicitly specified, in which case NODES will be used.

   CHILDREN = { NONE | SETS | CONTENTS }
   SETS     = { NONE | SETS | CONTENTS }

If reading only part of a file, specify whether or not child or contained sets (CHILDREN and SETS, respectively) of input sets are to be read. The options are:

  • NONE - Do not read sets because they are children of designated sets.
  • SETS - Read all child sets of designated input sets.
  • CONTENTS - (Default). Read all child sets and any entities contained in those sets.

BUFFER_SIZE=<BYTES>: Reduce buffer size for debugging purposes.

KEEP: For debugging purposes, retain partially written file if a failure occurs during the write process.

BLOCKED_COORDINATE_IO={yes|no}: During read of HDF5 file, read vertex coordinates in blocked format (read all X coordinates, followed by all Y coordinates, etc.) Default is 'no'.

BCAST_SUMMARY={yes|no}: During parallel read of HDF5 file, read file summary data on root process as serial IO and broadcast summary to other processes. All processes then re-open file for parallel IO. If 'no', file is opened only once by all processes for parallel IO and all processes read summary data. Default is 'yes'.

BCAST_DUPLICATE_READS={yes|no}: Work around MPI IO limitations when all processes read an indentical region of a file by reading the data only on the root process and using MPI_Bcast to send the data to other processes. Default is 'yes'.

HYPERSLAB_OR: During partial or parallel read, use documented (default) method for building HDF5 hyperslabs. This option is deprecated. Reader should automatically detect if HYPERSLAB_APPEND works and if not will default to HYPERSLAB_OR. This option is retaind for debugging purposes only.

HYPERSLAB_APPEND: During partial or parallel read using a modified (i.e. hacked) HDF5 library, utilize hack for more efficient hyperslab selection construction. This option is deprecated. Reader should automatically detect if HYPERSLAB_APPEND works and if not will default to HYPERSLAB_OR.

HYPERSLAB_SELECT_LIMIT=n: Set upper bound on the number of HDF5 hyperslabs that can be combined for a single read from an dataset. The default is defined by DEFAULT_HYPERSLAB_SELECT_LIMIT in ReadHDF5Dataset.cpp. If HYPERSLAB_APPEND is specified and this option is not, then the default is no limit. This limit should be removed for future HDF5 releases (> 1.8.x) as said future releases will no longer require the HYPERSLAB_APPEND hack in order to avoid O(n^2) hyperslab selection behavior.