Hi all. I have a question about the parallel writing with data chunks. As it is explained in the hdfgroup homepage( http://www.hdfgroup.org/HDF5/Tutor/phypechk.html) I tried the posted example with different dimension size.
1. In the example, dimension of the total data is 4 by 8 (in fortran) and there are 4 cores so each core shares 2 by 4 data chunks. If I increase 4 by 8 to 5 by 9, and assign corresponding data chunk size to each core (core1: 2X4, core2: 3X4, core3: 2X5, core4: 3X5), it runs but writes out in wrong format. 2. If I do not use "h5pset_chunk_f(plist_id, rank, chunk_dims, error)" subroutine call, then it works fine with 5 by 9 size. It prints out the result as I intended. 3. 4 by 8 example works also fine with datset created without 'plist_id' from "h5pset_chunk_f(plist_id, rank, chunk_dims, error)" subroutine call. My question is 1. To use "h5pset_chunk_f" subroutine, should the chunk size of each core has to be the same? 2. If parallel writing works with and without "h5pset_chunk_f" subroutine, what is the use of this subroutine? Sorry for the basic question but I could not find any reference that explains it clearly and easily. Thanks in advance. Seungbum
_______________________________________________ Hdf-forum is for HDF software users discussion. Hdf-forum@hdfgroup.org http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
