Hdf5DataWriter.hpp

/*

Copyright (c) 2005-2017, University of Oxford.
All rights reserved.

University of Oxford means the Chancellor, Masters and Scholars of the
University of Oxford, having an administrative office at Wellington
Square, Oxford OX1 2JD, UK.

This file is part of Chaste.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
 * Redistributions of source code must retain the above copyright notice,
   this list of conditions and the following disclaimer.
 * Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.
 * Neither the name of the University of Oxford nor the names of its
   contributors may be used to endorse or promote products derived from this
   software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

*/

#ifndef HDF5DATAWRITER_HPP_
#define HDF5DATAWRITER_HPP_

#include <vector>

#include "AbstractHdf5Access.hpp"
#include "DataWriterVariable.hpp"
#include "DistributedVectorFactory.hpp"

class Hdf5DataWriter : public AbstractHdf5Access //: public AbstractDataWriter
{
    friend class TestHdf5DataWriter;
private:

    DistributedVectorFactory& mrVectorFactory;

    const bool mCleanDirectory;                     
    const bool mUseExistingFile;                    
    bool mIsInDefineMode;                           
    bool mIsFixedDimensionSet;                      
    unsigned mEstimatedUnlimitedLength;             
    unsigned mFileFixedDimensionSize;               
    unsigned mDataFixedDimensionSize;               
    unsigned mLo;                                   
    unsigned mHi;                                   
    unsigned mNumberOwned;                          
    unsigned mOffset;                               
    bool mNeedExtend;                               
    bool mUseMatrixForIncompleteData;               
    std::vector<DataWriterVariable> mVariables;     
    long unsigned mCurrentTimeStep;                 
    Mat mSinglePermutation;                         
    Mat mDoublePermutation;                         
    Mat mSingleIncompleteOutputMatrix;              
    Mat mDoubleIncompleteOutputMatrix;              
    bool mUseOptimalChunkSizeAlgorithm;             
    hsize_t mChunkSize[DATASET_DIMS];               
    hsize_t mNumberOfChunks;                        
    hsize_t mFixedChunkSize[DATASET_DIMS];          
    hsize_t mChunkTargetSize;                       
    hsize_t mAlignment;                             
    bool mUseCache;                                 
    long unsigned mCacheFirstTimeStep;              
    std::vector<double> mDataCache;                 
    void CheckVariableName(const std::string& rName);

    void CheckUnitsName(const std::string& rName);

    void ComputeIncompleteOffset();

    void OpenFile();

    hsize_t CalculateNumberOfChunks();

    void CalculateChunkDims( unsigned targetSize, unsigned* pChunkSizeInBytes, bool* pAllOneChunk );

    void SetChunkSize();

public:

    Hdf5DataWriter(DistributedVectorFactory& rVectorFactory,
                   const std::string& rDirectory,
                   const std::string& rBaseName,
                   bool cleanDirectory=true,
                   bool extendData=false,
                   std::string datasetName="Data",
                   bool useCache=false);

    virtual ~Hdf5DataWriter();

    void DefineFixedDimension(long dimensionSize);

    void DefineFixedDimension(const std::vector<unsigned>& rNodesToOuput, long vecSize);

    void DefineUnlimitedDimension(const std::string& rVariableName, const std::string& rVariableUnits, unsigned estimatedLength = 1);

    void AdvanceAlongUnlimitedDimension();

    int DefineVariable(const std::string& rVariableName, const std::string& rVariableUnits);

    bool IsInDefineMode();

    virtual void EndDefineMode();

    void PossiblyExtend();

    void EmptyDataset();

    void PutVector(int variableID, Vec petscVector);

    void PutStripedVector(std::vector<int> variableIDs, Vec petscVector);

    bool GetUsingCache();

    void WriteCache();

    void PutUnlimitedVariable(double value);

    void Close();

    int GetVariableByName(const std::string& rVariableName);


    bool ApplyPermutation(const std::vector<unsigned>& rPermutation, bool unsafeExtendingMode=false);

    void DefineFixedDimensionUsingMatrix(const std::vector<unsigned>& rNodesToOuput, long vecSize);

    void SetFixedChunkSize(const unsigned& rTimestepsPerChunk,
                           const unsigned& rNodesPerChunk,
                           const unsigned& rVariablesPerChunk);

    /*
     * * NOTES ON CHUNK SIZE AND ALIGNMENT *
     *
     * The default target chunk size is 128 K, which seems to be a good compromise
     * for small problems (e.g. on a desktop PC). For larger problems, I/O
     * performance often improves with increased chunk size. A sweet spot seems to
     * be 1 M chunks.
     *
     * On a striped filesystem, for best performance set the chunk size and
     * alignment (using `H5Pset_alignment` above) to the file stripe size. With
     * `H5Pset_alignment`, every chunk starts at a multiple of the alignment value.
     *
     * To avoid wasting space, the chunk size should be an integer multiple of the
     * alignment value. Note that the algorithm below automatically goes back one
     * step after exceeding the chunk size, which minimises wasted space. To see
     * why, consider the examples below.
     *
     * (Example 1) Say our file system uses 1 M stripes. If we set
     *     target_size_in_bytes = 1024*1024;
     * below and uncomment
     *     H5Pset_alignment(fapl, 0, 1024*1024);
     * above, i.e. aim for (slightly under) 1 M chunks and align them to 1 M
     * boundaries, then the algorithm below will get as close as possible to 1 M
     * chunks but not exceed it, so each chunk will be padded slightly to sit on
     * the 1 M boundaries. Each chunk will therefore have its own stripe on the
     * file system, which should give us the best bandwidth and least contention.
     * Conclusion: this is optimal!
     *
     * Note: In general the algorithm can get very close to the target so the
     * waste isn't bad. Typical utilization is 99.99% (check with "h5ls -v ...").
     *
     * (Example 2) We set
     *     target_size_in_bytes = 128*1024;
     * and uncomment
     *     H5Pset_alignment(fapl, 0, 1024*1024);
     * i.e. 128 K chunks aligned to 1 M boundaries. This would pad every chunk to
     * 1 M boundaries, wasting 7/8 of the space in the file! A file which might be
     * 5 G with an efficient layout would be more like 40 G! Conclusion: setting
     * the chunk size to less than the alignment value is very bad!
     *
     * (Example 3) Say our file system uses 1 M stripes. We set
     *     target_size_in_bytes = 2*1024*1024;
     * and uncomment
     *     H5Pset_alignment(fapl, 0, 1024*1024);
     * i.e. 2 M chunks aligned to 1 M boundaries. This might not be optimal, but
     * it's OK, since the chunk size is (slightly under) twice the alignment, as in
     * Example 1 the amount of padding would be very small. Each read/write would
     * require accessing 2 stripes on the file system. Conclusion: a chunk size of
     * an integer multiple of the alignment value is fine (but not optimal).
     */

    void SetTargetChunkSize(hsize_t targetSize);

    void SetAlignment(hsize_t alignment);
};

#endif /*HDF5DATAWRITER_HPP_*/