Original commit message:
------------------------
Parallel IO: New collated file format
When an OpenFOAM simulation runs in parallel, the data for decomposed fields and
mesh(es) has historically been stored in multiple files within separate
directories for each processor. Processor directories are named 'processorN',
where N is the processor number.
This commit introduces an alternative "collated" file format where the data for
each decomposed field (and mesh) is collated into a single file, which is
written and read on the master processor. The files are stored in a single
directory named 'processors'.
The new format produces significantly fewer files - one per field, instead of N
per field. For large parallel cases, this avoids the restriction on the number
of open files imposed by the operating system limits.
The file writing can be threaded allowing the simulation to continue running
while the data is being written to file. NFS (Network File System) is not
needed when using the the collated format and additionally, there is an option
to run without NFS with the original uncollated approach, known as
"masterUncollated".
The controls for the file handling are in the OptimisationSwitches of
etc/controlDict:
OptimisationSwitches
{
...
//- Parallel IO file handler
// uncollated (default), collated or masterUncollated
fileHandler uncollated;
//- collated: thread buffer size for queued file writes.
// If set to 0 or not sufficient for the file size threading is not used.
// Default: 2e9
maxThreadFileBufferSize 2e9;
//- masterUncollated: non-blocking buffer size.
// If the file exceeds this buffer size scheduled transfer is used.
// Default: 2e9
maxMasterFileBufferSize 2e9;
}
When using the collated file handling, memory is allocated for the data in the
thread. maxThreadFileBufferSize sets the maximum size of memory in bytes that
is allocated. If the data exceeds this size, the write does not use threading.
When using the masterUncollated file handling, non-blocking MPI communication
requires a sufficiently large memory buffer on the master node.
maxMasterFileBufferSize sets the maximum size in bytes of the buffer. If the
data exceeds this size, the system uses scheduled communication.
The installation defaults for the fileHandler choice, maxThreadFileBufferSize
and maxMasterFileBufferSize (set in etc/controlDict) can be over-ridden within
the case controlDict file, like other parameters. Additionally the fileHandler
can be set by:
- the "-fileHandler" command line argument;
- a FOAM_FILEHANDLER environment variable.
A foamFormatConvert utility allows users to convert files between the collated
and uncollated formats, e.g.
mpirun -np 2 foamFormatConvert -parallel -fileHandler uncollated
An example case demonstrating the file handling methods is provided in:
$FOAM_TUTORIALS/IO/fileHandling
The work was undertaken by Mattijs Janssens, in collaboration with Henry Weller.
- use allocator class to wrap the stream pointers instead of passing
them into ISstream, OSstream and using a dynamic cast to delete
then. This is especially important if we will have a bidirectional
stream (can't delete twice!).
STYLE:
- file stream constructors with std::string (C++11)
- for rewind, explicit about in|out direction. This is not currently
important, but avoids surprises with any future bidirectional access.
- combined string streams in StringStream.H header.
Similar to <sstream> include that has both input and output string
streams.
- STLpoint.H
- isoAdvection.C
- checkMesh/writeFields.C
STYLE: drop construct STLpoint(Istream&), since it doesn't make much sense
- No use case for reading via an OpenFOAM stream and tokenizer.
Should always be parsing ASCII or reading binary directly.
- this shifts responsibility away from caller to the individual writers
for knowing which file formats are supported and which file ending is
appropriate. When the writer receives the output format request,
it can elect to downgrade or otherwise adjust it to what it can
actually manage (eg, legacy vs xml vs xml-append).
But currently still just with legacy format backends.
Adds overset discretisation to selected physics:
- diffusion : overLaplacianDyMFoam
- incompressible steady : overSimpleFoam
- incompressible transient : overPimpleDyMFoam
- compressible transient: overRhoPimpleDyMFoam
- two-phase VOF: overInterDyMFoam
The overset method chosen is a parallel, fully implicit implementation
whereby the interpolation (from donor to acceptor) is inserted as an
adapted discretisation on the donor cells, such that the resulting matrix
can be solved using the standard linear solvers.
Above solvers come with a set of tutorials, showing how to create and set-up
simple simulations from scratch.
- Constructor for bounding box of a single point.
- add(boundBox), add(point) ...
-> Extend box to enclose the second box or point(s).
Eg,
bb.add(pt);
vs.
bb.min() = Foam::min(bb.min(), pt);
bb.max() = Foam::max(bb.max(), pt);
Also works with other bounding boxes.
Eg,
bb.add(bb2);
// OR
bb += bb2;
vs.
bb.min() = Foam::min(bb.min(), bb2.min());
bb.max() = Foam::max(bb.max(), bb2.max());
'+=' operator allows the reduction to be used in parallel
gather/scatter operations.
A global '+' operator is not currently needed.
Note: may be useful in the future to have a 'clear()' method
that resets to a zero-sized (inverted) box.
STYLE: make many bounding box constructors explicit
reduce()
- parallel reduction of min/max values.
Reduces coding for the callers.
Eg,
bb.reduce();
instead of the previous method:
reduce(bb.min(), minOp<point>());
reduce(bb.max(), maxOp<point>());
STYLE:
- use initializer list for creating static content
- use point::min/point::max when defining standard boxes
cellZones and pointZones can now be created in one action without the
need to first create a cellSet or pointSet and converting that to the
corresponding zone, e.g.
actions
(
// Example: create cellZone from a box region
{
name c0;
type cellZoneSet;
action new;
source boxToCell;
sourceInfo
{
box (0.04 0 0)(0.06 100 100);
}
}
);
On 64-bit systems, the system installations of boost, cgal are under
lib64/. The behaviour for a ThirdParty build is mostly lib/ but this
can also be changing.
Boost 1_62_0 and older build into 'lib/'.
CGAL-4.9 builds into 'lib64/', older versions into 'lib/'.
Future-proof things by using lib$WM_COMPILER_LIB_ARCH for boost and
cgal build rules, and forcing these as build targets in the ThirdParty
makeCGAL as well.
--
STYLE: check for boost/version.hpp, CGAL/version.h instead their directories
- Place common code under OSspecific.
By including "endian.H", either one of WM_BIG_ENDIAN or WM_LITTLE_ENDIAN
will be defined.
Provides inline 32-bit and 64-bit byte swap routines that can be
used/re-used elsewhere.
The inplace memory swaps currently used by the VTK output are left for
the moment pending further cleanup of that code.
- Allows passing of additional information (per-face zone ids) or possibly
other things, while reducing the number of arguments to pass.
- In sampledTriSurfaceMesh, preserve the region information that was
read in, passing it onwards via the UnsortedMeshSurface content.
The Nastran surface writer is currently the only writer making use
of this per-face zone information.
Passing it through as a PSHELL attribute, which should retain the
distinction for parts. (issue #204)
- the checking for point-connected multiple-regions now also writes the
conflicting points to a pointSet
- with the -writeSets option it now also reconstructs & writes pointSets
In parallel the sets are reconstructed. e.g.
mpirun -np 6 checkMesh -parallel -allGeometry -allTopology -writeSets vtk
will create a postProcessing/ folder with the vtk files of the
(reconstructed) faceSets and cellSets.
Also improved analysis of disconnected regions now also checks for point
connectivity with is useful for detecting if AMI regions have duplicate
points.
Patch contributed by Mattijs Janssens
splitMeshRegions: handle flipping of faces for surface fields
subsetMesh: subset dimensionedFields
decomposePar: use run-time selection of decomposition constraints. Used to
keep cells on particular processors. See the decomposeParDict in
$FOAM_UTILITIES/parallel/decomposePar:
- preserveBaffles: keep baffle faces on same processor
- preserveFaceZones: keep faceZones owner and neighbour on same processor
- preservePatches: keep owner and neighbour on same processor. Note: not
suitable for cyclicAMI since these are not coupled on the patch level
- singleProcessorFaceSets: keep complete faceSet on a single processor
- refinementHistory: keep cells originating from a single cell on the
same processor.
decomposePar: clean up decomposition of refinement data from snappyHexMesh
reconstructPar: reconstruct refinement data (refineHexMesh, snappyHexMesh)
reconstructParMesh: reconstruct refinement data (refineHexMesh, snappyHexMesh)
redistributePar:
- corrected mapping surfaceFields
- adding processor patches in order consistent with decomposePar
argList: check that slaves are running same version as master
fvMeshSubset: move to dynamicMesh library
fvMeshDistribute:
- support for mapping dimensionedFields
- corrected mapping of surfaceFields
parallel routines: allow parallel running on single processor
Field: support for
- distributed mapping
- mapping with flipping
mapDistribute: support for flipping
AMIInterpolation: avoid constructing localPoints
to have the prefix 'write' rather than 'output'
So outputTime() -> writeTime()
but 'outputTime()' is still supported for backward-compatibility.
Also removed the redundant secondary-writing functionality from Time
which has been superseded by the 'writeRegisteredObject' functionObject.
These new names are more consistent and logical because:
primitiveField():
primitiveFieldRef():
Provides low-level access to the Field<Type> (primitive field)
without dimension or mesh-consistency checking. This should only be
used in the low-level functions where dimensional consistency is
ensured by careful programming and computational efficiency is
paramount.
internalField():
internalFieldRef():
Provides access to the DimensionedField<Type, GeoMesh> of values on
the internal mesh-type for which the GeometricField is defined and
supports dimension and checking and mesh-consistency checking.
The deprecated non-const tmp functionality is now on the compiler switch
NON_CONST_TMP which can be enabled by adding -DNON_CONST_TMP to EXE_INC
in the Make/options file. However, it is recommended to upgrade all
code to the new safer tmp by using the '.ref()' member function rather
than the non-const '()' dereference operator when non-const access to
the temporary object is required.
Please report any problems on Mantis.
Henry G. Weller
CFD Direct.
Moved file path handling to regIOobject and made it type specific so
now every object can have its own rules. Examples:
- faceZones are now processor local (and don't search up anymore)
- timeStampMaster is now no longer hardcoded inside IOdictionary
(e.g. uniformDimensionedFields support it as well)
- the distributedTriSurfaceMesh is properly processor-local; no need
for fileModificationChecking manipulation.
- redistributePar to have almost (complete) functionality of decomposePar+reconstructPar
- low-level distributed Field mapping
- support for mapping surfaceFields (including flipping faces)
- support for decomposing/reconstructing refinement data
- checkMesh has option to write faceSets or (outside of) cellSets as
sampledSurface format. It automatically reconstructs the set on the master
and writes it to the postProcessing folder (as any sampledSurface). E.g.
mpirun -np 6 checkMesh -allTopology -allGeometry -writeSets vtk -parallel
- fixed order writing of symmTensor in Ensight writers
The built-in explicit symplectic integrator has been replaced by a
general framework supporting run-time selectable integrators. Currently
the explicit symplectic, implicit Crank-Nicolson and implicit Newmark
methods are provided, all of which are 2nd-order in time:
Symplectic 2nd-order explicit time-integrator for 6DoF solid-body motion:
Reference:
Dullweber, A., Leimkuhler, B., & McLachlan, R. (1997).
Symplectic splitting methods for rigid body molecular dynamics.
The Journal of chemical physics, 107(15), 5840-5851.
Can only be used for explicit integration of the motion of the body,
i.e. may only be called once per time-step, no outer-correctors may be
applied. For implicit integration with outer-correctors choose either
CrankNicolson or Newmark schemes.
Example specification in dynamicMeshDict:
solver
{
type symplectic;
}
Newmark 2nd-order time-integrator for 6DoF solid-body motion:
Reference:
Newmark, N. M. (1959).
A method of computation for structural dynamics.
Journal of the Engineering Mechanics Division, 85(3), 67-94.
Example specification in dynamicMeshDict:
solver
{
type Newmark;
gamma 0.5; // Velocity integration coefficient
beta 0.25; // Position integration coefficient
}
Crank-Nicolson 2nd-order time-integrator for 6DoF solid-body motion:
The off-centering coefficients for acceleration (velocity integration) and
velocity (position/orientation integration) may be specified but default
values of 0.5 for each are used if they are not specified. With the default
off-centering this scheme is equivalent to the Newmark scheme with default
coefficients.
Example specification in dynamicMeshDict:
solver
{
type CrankNicolson;
aoc 0.5; // Acceleration off-centering coefficient
voc 0.5; // Velocity off-centering coefficient
}
Both the Newmark and Crank-Nicolson are proving more robust and reliable
than the symplectic method for solving complex coupled problems and the
tutorial cases have been updated to utilize this.
In this new framework it would be straight forward to add other methods
should the need arise.
Henry G. Weller
CFD Direct
Command-line option handling:
+ If -all specified or no refineMeshDict exists or, refine all cells
+ If -dict <file> specified refine according to <file>
+ If refineMeshDict exists refine according to refineMeshDict
When the refinement or all cells is selected apply 3D refinement for 3D
cases and 2D refinement for 2D cases.
To compile with 64bit labels set
WM_LABEL_SIZE=64
in ~/OpenFOAM/dev/prefs.sh
source ~/.bashrc
then Allwmake in OpenFOAM-dev.
This will build into for example OpenFOAM-dev/platforms/linux64ClangDPInt64Opt
If WM_LABEL_SIZE is unset or set to 32:
WM_LABEL_SIZE=32
the build would be placed into OpenFOAM-dev/platforms/linux64ClangDPInt32Opt
Thus both 32bit and 64bit label builds can coexist without problem.
