openfoam/applications/utilities/parallelProcessing/decomposePar/domainDecomposition.C
Andrew Heather d8d6030ab6 INT: Integration of Mattijs' collocated parallel IO additions
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.
2017-07-07 11:39:56 +01:00

977 lines
28 KiB
C

/*---------------------------------------------------------------------------*\
========= |
\\ / F ield | OpenFOAM: The Open Source CFD Toolbox
\\ / O peration |
\\ / A nd | Copyright (C) 2011-2016 OpenFOAM Foundation
\\/ M anipulation |
-------------------------------------------------------------------------------
License
This file is part of OpenFOAM.
OpenFOAM is free software: you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
OpenFOAM is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License
along with OpenFOAM. If not, see <http://www.gnu.org/licenses/>.
\*---------------------------------------------------------------------------*/
#include "domainDecomposition.H"
#include "dictionary.H"
#include "labelIOList.H"
#include "processorPolyPatch.H"
#include "processorCyclicPolyPatch.H"
#include "fvMesh.H"
#include "OSspecific.H"
#include "Map.H"
#include "DynamicList.H"
#include "fvFieldDecomposer.H"
#include "IOobjectList.H"
#include "cellSet.H"
#include "faceSet.H"
#include "pointSet.H"
#include "decompositionModel.H"
#include "hexRef8Data.H"
// * * * * * * * * * * * * * Private Member Functions * * * * * * * * * * * //
void Foam::domainDecomposition::mark
(
const labelList& zoneElems,
const label zoneI,
labelList& elementToZone
)
{
forAll(zoneElems, i)
{
label pointi = zoneElems[i];
if (elementToZone[pointi] == -1)
{
// First occurrence
elementToZone[pointi] = zoneI;
}
else if (elementToZone[pointi] >= 0)
{
// Multiple zones
elementToZone[pointi] = -2;
}
}
}
// * * * * * * * * * * * * * * * * Constructors * * * * * * * * * * * * * * //
// from components
Foam::domainDecomposition::domainDecomposition
(
const IOobject& io,
const fileName& decompDictFile
)
:
fvMesh(io),
facesInstancePointsPtr_
(
pointsInstance() != facesInstance()
? new pointIOField
(
IOobject
(
"points",
facesInstance(),
polyMesh::meshSubDir,
*this,
IOobject::MUST_READ,
IOobject::NO_WRITE,
false
)
)
: nullptr
),
decompDictFile_(decompDictFile),
nProcs_
(
readInt
(
decompositionModel::New
(
*this,
decompDictFile
).lookup("numberOfSubdomains")
)
),
distributed_(false),
cellToProc_(nCells()),
procPointAddressing_(nProcs_),
procFaceAddressing_(nProcs_),
procCellAddressing_(nProcs_),
procPatchSize_(nProcs_),
procPatchStartIndex_(nProcs_),
procNeighbourProcessors_(nProcs_),
procProcessorPatchSize_(nProcs_),
procProcessorPatchStartIndex_(nProcs_),
procProcessorPatchSubPatchIDs_(nProcs_),
procProcessorPatchSubPatchStarts_(nProcs_)
{
decompositionModel::New
(
*this,
decompDictFile
).readIfPresent("distributed", distributed_);
}
// * * * * * * * * * * * * * * * * Destructor * * * * * * * * * * * * * * * //
Foam::domainDecomposition::~domainDecomposition()
{}
// * * * * * * * * * * * * * * * Member Functions * * * * * * * * * * * * * //
bool Foam::domainDecomposition::writeDecomposition(const bool decomposeSets)
{
Info<< "\nConstructing processor meshes" << endl;
// Mark point/faces/cells that are in zones.
// -1 : not in zone
// -2 : in multiple zones
// >= 0 : in single given zone
// This will give direct lookup of elements that are in a single zone
// and we'll only have to revert back to searching through all zones
// for the duplicate elements
// Point zones
labelList pointToZone(points().size(), -1);
forAll(pointZones(), zonei)
{
mark(pointZones()[zonei], zonei, pointToZone);
}
// Face zones
labelList faceToZone(faces().size(), -1);
forAll(faceZones(), zonei)
{
mark(faceZones()[zonei], zonei, faceToZone);
}
// Cell zones
labelList cellToZone(nCells(), -1);
forAll(cellZones(), zonei)
{
mark(cellZones()[zonei], zonei, cellToZone);
}
PtrList<const cellSet> cellSets;
PtrList<const faceSet> faceSets;
PtrList<const pointSet> pointSets;
if (decomposeSets)
{
// Read sets
IOobjectList objects(*this, facesInstance(), "polyMesh/sets");
{
IOobjectList cSets(objects.lookupClass(cellSet::typeName));
forAllConstIter(IOobjectList, cSets, iter)
{
cellSets.append(new cellSet(*iter()));
}
}
{
IOobjectList fSets(objects.lookupClass(faceSet::typeName));
forAllConstIter(IOobjectList, fSets, iter)
{
faceSets.append(new faceSet(*iter()));
}
}
{
IOobjectList pSets(objects.lookupClass(pointSet::typeName));
forAllConstIter(IOobjectList, pSets, iter)
{
pointSets.append(new pointSet(*iter()));
}
}
}
// Load refinement data (if any)
hexRef8Data baseMeshData
(
IOobject
(
"dummy",
facesInstance(),
polyMesh::meshSubDir,
*this,
IOobject::READ_IF_PRESENT,
IOobject::NO_WRITE,
false
)
);
label maxProcCells = 0;
label totProcFaces = 0;
label maxProcPatches = 0;
label totProcPatches = 0;
label maxProcFaces = 0;
// Write out the meshes
for (label proci = 0; proci < nProcs_; proci++)
{
// Create processor points
const labelList& curPointLabels = procPointAddressing_[proci];
const pointField& meshPoints = points();
labelList pointLookup(nPoints(), -1);
pointField procPoints(curPointLabels.size());
forAll(curPointLabels, pointi)
{
procPoints[pointi] = meshPoints[curPointLabels[pointi]];
pointLookup[curPointLabels[pointi]] = pointi;
}
// Create processor faces
const labelList& curFaceLabels = procFaceAddressing_[proci];
const faceList& meshFaces = faces();
labelList faceLookup(nFaces(), -1);
faceList procFaces(curFaceLabels.size());
forAll(curFaceLabels, facei)
{
// Mark the original face as used
// Remember to decrement the index by one (turning index)
//
label curF = mag(curFaceLabels[facei]) - 1;
faceLookup[curF] = facei;
// get the original face
labelList origFaceLabels;
if (curFaceLabels[facei] >= 0)
{
// face not turned
origFaceLabels = meshFaces[curF];
}
else
{
origFaceLabels = meshFaces[curF].reverseFace();
}
// translate face labels into local point list
face& procFaceLabels = procFaces[facei];
procFaceLabels.setSize(origFaceLabels.size());
forAll(origFaceLabels, pointi)
{
procFaceLabels[pointi] = pointLookup[origFaceLabels[pointi]];
}
}
// Create processor cells
const labelList& curCellLabels = procCellAddressing_[proci];
const cellList& meshCells = cells();
cellList procCells(curCellLabels.size());
forAll(curCellLabels, celli)
{
const labelList& origCellLabels = meshCells[curCellLabels[celli]];
cell& curCell = procCells[celli];
curCell.setSize(origCellLabels.size());
forAll(origCellLabels, cellFacei)
{
curCell[cellFacei] = faceLookup[origCellLabels[cellFacei]];
}
}
// Create processor mesh without a boundary
fileName processorCasePath
(
time().caseName()/fileName(word("processor") + Foam::name(proci))
);
// create a database
Time processorDb
(
Time::controlDictName,
time().rootPath(),
processorCasePath,
word("system"),
word("constant")
);
processorDb.setTime(time());
// create the mesh. Two situations:
// - points and faces come from the same time ('instance'). The mesh
// will get constructed in the same instance.
// - points come from a different time (moving mesh cases).
// It will read the points belonging to the faces instance and
// construct the procMesh with it which then gets handled as above.
// (so with 'old' geometry).
// Only at writing time will it additionally write the current
// points.
autoPtr<polyMesh> procMeshPtr;
if (facesInstancePointsPtr_.valid())
{
// Construct mesh from facesInstance.
pointField facesInstancePoints
(
facesInstancePointsPtr_(),
curPointLabels
);
procMeshPtr.reset
(
new polyMesh
(
IOobject
(
this->polyMesh::name(), // region of undecomposed mesh
facesInstance(),
processorDb
),
xferMove(facesInstancePoints),
xferMove(procFaces),
xferMove(procCells)
)
);
}
else
{
procMeshPtr.reset
(
new polyMesh
(
IOobject
(
this->polyMesh::name(), // region of undecomposed mesh
facesInstance(),
processorDb
),
xferMove(procPoints),
xferMove(procFaces),
xferMove(procCells)
)
);
}
polyMesh& procMesh = procMeshPtr();
// Create processor boundary patches
const labelList& curPatchSizes = procPatchSize_[proci];
const labelList& curPatchStarts = procPatchStartIndex_[proci];
const labelList& curNeighbourProcessors =
procNeighbourProcessors_[proci];
const labelList& curProcessorPatchSizes =
procProcessorPatchSize_[proci];
const labelList& curProcessorPatchStarts =
procProcessorPatchStartIndex_[proci];
const labelListList& curSubPatchIDs =
procProcessorPatchSubPatchIDs_[proci];
const labelListList& curSubStarts =
procProcessorPatchSubPatchStarts_[proci];
const polyPatchList& meshPatches = boundaryMesh();
// Count the number of inter-proc patches
label nInterProcPatches = 0;
forAll(curSubPatchIDs, procPatchi)
{
nInterProcPatches += curSubPatchIDs[procPatchi].size();
}
List<polyPatch*> procPatches
(
curPatchSizes.size() + nInterProcPatches,
reinterpret_cast<polyPatch*>(0)
);
label nPatches = 0;
forAll(curPatchSizes, patchi)
{
// Get the face labels consistent with the field mapping
// (reuse the patch field mappers)
const polyPatch& meshPatch = meshPatches[patchi];
fvFieldDecomposer::patchFieldDecomposer patchMapper
(
SubList<label>
(
curFaceLabels,
curPatchSizes[patchi],
curPatchStarts[patchi]
),
meshPatch.start()
);
// Map existing patches
procPatches[nPatches] = meshPatch.clone
(
procMesh.boundaryMesh(),
nPatches,
patchMapper.directAddressing(),
curPatchStarts[patchi]
).ptr();
nPatches++;
}
forAll(curProcessorPatchSizes, procPatchi)
{
const labelList& subPatchID = curSubPatchIDs[procPatchi];
const labelList& subStarts = curSubStarts[procPatchi];
label curStart = curProcessorPatchStarts[procPatchi];
forAll(subPatchID, i)
{
label size =
(
i < subPatchID.size()-1
? subStarts[i+1] - subStarts[i]
: curProcessorPatchSizes[procPatchi] - subStarts[i]
);
if (subPatchID[i] == -1)
{
// From internal faces
procPatches[nPatches] =
new processorPolyPatch
(
size,
curStart,
nPatches,
procMesh.boundaryMesh(),
proci,
curNeighbourProcessors[procPatchi]
);
}
else
{
const coupledPolyPatch& pcPatch
= refCast<const coupledPolyPatch>
(
boundaryMesh()[subPatchID[i]]
);
procPatches[nPatches] =
new processorCyclicPolyPatch
(
size,
curStart,
nPatches,
procMesh.boundaryMesh(),
proci,
curNeighbourProcessors[procPatchi],
pcPatch.name(),
pcPatch.transform()
);
}
curStart += size;
nPatches++;
}
}
// Add boundary patches
procMesh.addPatches(procPatches);
// Create and add zones
// Point zones
{
const pointZoneMesh& pz = pointZones();
// Go through all the zoned points and find out if they
// belong to a zone. If so, add it to the zone as
// necessary
List<DynamicList<label>> zonePoints(pz.size());
// Estimate size
forAll(zonePoints, zonei)
{
zonePoints[zonei].setCapacity(pz[zonei].size()/nProcs_);
}
// Use the pointToZone map to find out the single zone (if any),
// use slow search only for shared points.
forAll(curPointLabels, pointi)
{
label curPoint = curPointLabels[pointi];
label zonei = pointToZone[curPoint];
if (zonei >= 0)
{
// Single zone.
zonePoints[zonei].append(pointi);
}
else if (zonei == -2)
{
// Multiple zones. Lookup.
forAll(pz, zonei)
{
label index = pz[zonei].whichPoint(curPoint);
if (index != -1)
{
zonePoints[zonei].append(pointi);
}
}
}
}
procMesh.pointZones().clearAddressing();
procMesh.pointZones().setSize(zonePoints.size());
forAll(zonePoints, zonei)
{
procMesh.pointZones().set
(
zonei,
pz[zonei].clone
(
procMesh.pointZones(),
zonei,
zonePoints[zonei].shrink()
)
);
}
if (pz.size())
{
// Force writing on all processors
procMesh.pointZones().writeOpt() = IOobject::AUTO_WRITE;
}
}
// Face zones
{
const faceZoneMesh& fz = faceZones();
// Go through all the zoned face and find out if they
// belong to a zone. If so, add it to the zone as
// necessary
List<DynamicList<label>> zoneFaces(fz.size());
List<DynamicList<bool>> zoneFaceFlips(fz.size());
// Estimate size
forAll(zoneFaces, zonei)
{
label procSize = fz[zonei].size() / nProcs_;
zoneFaces[zonei].setCapacity(procSize);
zoneFaceFlips[zonei].setCapacity(procSize);
}
// Go through all the zoned faces and find out if they
// belong to a zone. If so, add it to the zone as
// necessary
forAll(curFaceLabels, facei)
{
// Remember to decrement the index by one (turning index)
//
label curF = mag(curFaceLabels[facei]) - 1;
label zonei = faceToZone[curF];
if (zonei >= 0)
{
// Single zone. Add the face
zoneFaces[zonei].append(facei);
label index = fz[zonei].whichFace(curF);
bool flip = fz[zonei].flipMap()[index];
if (curFaceLabels[facei] < 0)
{
flip = !flip;
}
zoneFaceFlips[zonei].append(flip);
}
else if (zonei == -2)
{
// Multiple zones. Lookup.
forAll(fz, zonei)
{
label index = fz[zonei].whichFace(curF);
if (index != -1)
{
zoneFaces[zonei].append(facei);
bool flip = fz[zonei].flipMap()[index];
if (curFaceLabels[facei] < 0)
{
flip = !flip;
}
zoneFaceFlips[zonei].append(flip);
}
}
}
}
procMesh.faceZones().clearAddressing();
procMesh.faceZones().setSize(zoneFaces.size());
forAll(zoneFaces, zonei)
{
procMesh.faceZones().set
(
zonei,
fz[zonei].clone
(
zoneFaces[zonei].shrink(), // addressing
zoneFaceFlips[zonei].shrink(), // flipmap
zonei,
procMesh.faceZones()
)
);
}
if (fz.size())
{
// Force writing on all processors
procMesh.faceZones().writeOpt() = IOobject::AUTO_WRITE;
}
}
// Cell zones
{
const cellZoneMesh& cz = cellZones();
// Go through all the zoned cells and find out if they
// belong to a zone. If so, add it to the zone as
// necessary
List<DynamicList<label>> zoneCells(cz.size());
// Estimate size
forAll(zoneCells, zonei)
{
zoneCells[zonei].setCapacity(cz[zonei].size()/nProcs_);
}
forAll(curCellLabels, celli)
{
label curCelli = curCellLabels[celli];
label zonei = cellToZone[curCelli];
if (zonei >= 0)
{
// Single zone.
zoneCells[zonei].append(celli);
}
else if (zonei == -2)
{
// Multiple zones. Lookup.
forAll(cz, zonei)
{
label index = cz[zonei].whichCell(curCelli);
if (index != -1)
{
zoneCells[zonei].append(celli);
}
}
}
}
procMesh.cellZones().clearAddressing();
procMesh.cellZones().setSize(zoneCells.size());
forAll(zoneCells, zonei)
{
procMesh.cellZones().set
(
zonei,
cz[zonei].clone
(
zoneCells[zonei].shrink(),
zonei,
procMesh.cellZones()
)
);
}
if (cz.size())
{
// Force writing on all processors
procMesh.cellZones().writeOpt() = IOobject::AUTO_WRITE;
}
}
// Set the precision of the points data to be min 10
IOstream::defaultPrecision(max(10u, IOstream::defaultPrecision()));
procMesh.write();
// Write points if pointsInstance differing from facesInstance
if (facesInstancePointsPtr_.valid())
{
pointIOField pointsInstancePoints
(
IOobject
(
"points",
pointsInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false
),
xferMove(procPoints)
);
pointsInstancePoints.write();
}
// Decompose any sets
if (decomposeSets)
{
forAll(cellSets, i)
{
const cellSet& cs = cellSets[i];
cellSet set(procMesh, cs.name(), cs.size()/nProcs_);
forAll(curCellLabels, i)
{
if (cs.found(curCellLabels[i]))
{
set.insert(i);
}
}
set.write();
}
forAll(faceSets, i)
{
const faceSet& cs = faceSets[i];
faceSet set(procMesh, cs.name(), cs.size()/nProcs_);
forAll(curFaceLabels, i)
{
if (cs.found(mag(curFaceLabels[i])-1))
{
set.insert(i);
}
}
set.write();
}
forAll(pointSets, i)
{
const pointSet& cs = pointSets[i];
pointSet set(procMesh, cs.name(), cs.size()/nProcs_);
forAll(curPointLabels, i)
{
if (cs.found(curPointLabels[i]))
{
set.insert(i);
}
}
set.write();
}
}
// Optional hexRef8 data
hexRef8Data
(
IOobject
(
"dummy",
facesInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false
),
baseMeshData,
procCellAddressing_[proci],
procPointAddressing_[proci]
).write();
// Statistics
Info<< endl
<< "Processor " << proci << nl
<< " Number of cells = " << procMesh.nCells()
<< endl;
maxProcCells = max(maxProcCells, procMesh.nCells());
label nBoundaryFaces = 0;
label nProcPatches = 0;
label nProcFaces = 0;
forAll(procMesh.boundaryMesh(), patchi)
{
if (isA<processorPolyPatch>(procMesh.boundaryMesh()[patchi]))
{
const processorPolyPatch& ppp =
refCast<const processorPolyPatch>
(
procMesh.boundaryMesh()[patchi]
);
Info<< " Number of faces shared with processor "
<< ppp.neighbProcNo() << " = " << ppp.size() << endl;
nProcPatches++;
nProcFaces += ppp.size();
}
else
{
nBoundaryFaces += procMesh.boundaryMesh()[patchi].size();
}
}
Info<< " Number of processor patches = " << nProcPatches << nl
<< " Number of processor faces = " << nProcFaces << nl
<< " Number of boundary faces = " << nBoundaryFaces << endl;
totProcFaces += nProcFaces;
totProcPatches += nProcPatches;
maxProcPatches = max(maxProcPatches, nProcPatches);
maxProcFaces = max(maxProcFaces, nProcFaces);
// create and write the addressing information
labelIOList pointProcAddressing
(
IOobject
(
"pointProcAddressing",
procMesh.facesInstance(),
procMesh.meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE
),
procPointAddressing_[proci]
);
pointProcAddressing.write();
labelIOList faceProcAddressing
(
IOobject
(
"faceProcAddressing",
procMesh.facesInstance(),
procMesh.meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE
),
procFaceAddressing_[proci]
);
faceProcAddressing.write();
labelIOList cellProcAddressing
(
IOobject
(
"cellProcAddressing",
procMesh.facesInstance(),
procMesh.meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE
),
procCellAddressing_[proci]
);
cellProcAddressing.write();
// Write patch map for backwards compatibility.
// (= identity map for original patches, -1 for processor patches)
label nMeshPatches = curPatchSizes.size();
labelList procBoundaryAddressing(identity(nMeshPatches));
procBoundaryAddressing.setSize(nMeshPatches+nProcPatches, -1);
labelIOList boundaryProcAddressing
(
IOobject
(
"boundaryProcAddressing",
procMesh.facesInstance(),
procMesh.meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE
),
procBoundaryAddressing
);
boundaryProcAddressing.write();
}
scalar avgProcCells = scalar(nCells())/nProcs_;
scalar avgProcPatches = scalar(totProcPatches)/nProcs_;
scalar avgProcFaces = scalar(totProcFaces)/nProcs_;
// In case of all faces on one processor. Just to avoid division by 0.
if (totProcPatches == 0)
{
avgProcPatches = 1;
}
if (totProcFaces == 0)
{
avgProcFaces = 1;
}
Info<< nl
<< "Number of processor faces = " << totProcFaces/2 << nl
<< "Max number of cells = " << maxProcCells
<< " (" << 100.0*(maxProcCells-avgProcCells)/avgProcCells
<< "% above average " << avgProcCells << ")" << nl
<< "Max number of processor patches = " << maxProcPatches
<< " (" << 100.0*(maxProcPatches-avgProcPatches)/avgProcPatches
<< "% above average " << avgProcPatches << ")" << nl
<< "Max number of faces between processors = " << maxProcFaces
<< " (" << 100.0*(maxProcFaces-avgProcFaces)/avgProcFaces
<< "% above average " << avgProcFaces << ")" << nl
<< endl;
return true;
}
// ************************************************************************* //