change chunk default size to 10MB (#925)

Co-authored-by: Ryan Ly <rly@lbl.gov>

src/hdmf/backends/hdf5/h5tools.py

@@ -29,6 +29,8 @@
 H5_REF = special_dtype(ref=Reference)
 H5_REGREF = special_dtype(ref=RegionReference)
 
+RDCC_NBYTES = 32*2**20  # set raw data chunk cache size = 32 MiB
+
 H5PY_3 = h5py.__version__.startswith('3')
 
 
@@ -745,7 +747,7 @@ def __read_ref(self, h5obj):
     def open(self):
         if self.__file is None:
             open_flag = self.__mode
-            kwargs = dict()
+            kwargs = dict(rdcc_nbytes=RDCC_NBYTES)
             if self.comm:
                 kwargs.update(driver='mpio', comm=self.comm)
 

src/hdmf/data_utils.py

@@ -154,7 +154,7 @@ class GenericDataChunkIterator(AbstractDataChunkIterator):
             doc=(
                 "If chunk_shape is not specified, it will be inferred as the smallest chunk "
                 "below the chunk_mb threshold.",
-                "Defaults to 1MB.",
+                "Defaults to 10MB.",
             ),
             default=None,
         ),
@@ -187,9 +187,8 @@ def __init__(self, **kwargs):
         Advanced users are offered full control over the shape parameters for the buffer and the chunks; however,
         the chunk shape must perfectly divide the buffer shape along each axis.
 
-        HDF5 also recommends not setting chunk_mb greater than 1 MB for optimal caching speeds.
-        See https://support.hdfgroup.org/HDF5/doc/TechNotes/TechNote-HDF5-ImprovingIOPerformanceCompressedDatasets.pdf
-        for more details.
+        HDF5 recommends chunk size in the range of 2 to 16 MB for optimal cloud performance.
+        https://youtu.be/rcS5vt-mKok?t=621
         """
         buffer_gb, buffer_shape, chunk_mb, chunk_shape, self.display_progress, self.progress_bar_options = getargs(
             "buffer_gb", "buffer_shape", "chunk_mb", "chunk_shape", "display_progress", "progress_bar_options", kwargs
@@ -198,7 +197,7 @@ def __init__(self, **kwargs):
         if buffer_gb is None and buffer_shape is None:
             buffer_gb = 1.0
         if chunk_mb is None and chunk_shape is None:
-            chunk_mb = 1.0
+            chunk_mb = 10.0
         assert (buffer_gb is not None) != (
             buffer_shape is not None
         ), "Only one of 'buffer_gb' or 'buffer_shape' can be specified!"

tests/unit/utils_test/test_core_GenericDataChunkIterator.py

@@ -277,17 +277,17 @@ def test_numpy_array_chunk_iterator(self):
 
     def test_buffer_shape_option(self):
         expected_buffer_shape = (1580, 316)
-        iterator_options = dict(buffer_shape=expected_buffer_shape)
+        iterator_options = dict(buffer_shape=expected_buffer_shape, chunk_mb=1.0)
         self.check_first_data_chunk_call(
             expected_selection=tuple([slice(0, buffer_shape_axis) for buffer_shape_axis in expected_buffer_shape]),
             iterator_options=iterator_options,
         )
         self.check_direct_hdf5_write(iterator_options=iterator_options)
 
     def test_buffer_gb_option(self):
-        # buffer is smaller than default chunk; should collapse to chunk shape
+        # buffer is smaller than chunk; should collapse to chunk shape
         resulting_buffer_shape = (1580, 316)
-        iterator_options = dict(buffer_gb=0.0005)
+        iterator_options = dict(buffer_gb=0.0005, chunk_mb=1.0)
         self.check_first_data_chunk_call(
             expected_selection=tuple(
                 [
@@ -334,14 +334,14 @@ def test_chunk_mb_option_while_condition(self):
         """Test to evoke while condition of default shaping method."""
         expected_chunk_shape = (2, 79, 79)
         special_array = np.random.randint(low=-(2 ** 15), high=2 ** 15 - 1, size=(2, 2000, 2000), dtype="int16")
-        iterator = self.TestNumpyArrayDataChunkIterator(array=special_array)
+        iterator = self.TestNumpyArrayDataChunkIterator(array=special_array, chunk_mb=1.0)
         self.assertEqual(iterator.chunk_shape, expected_chunk_shape)
 
     def test_chunk_mb_option_while_condition_unit_maxshape_axis(self):
         """Test to evoke while condition of default shaping method."""
         expected_chunk_shape = (1, 79, 79)
         special_array = np.random.randint(low=-(2 ** 15), high=2 ** 15 - 1, size=(1, 2000, 2000), dtype="int16")
-        iterator = self.TestNumpyArrayDataChunkIterator(array=special_array)
+        iterator = self.TestNumpyArrayDataChunkIterator(array=special_array, chunk_mb=1.0)
         self.assertEqual(iterator.chunk_shape, expected_chunk_shape)
 
     @unittest.skipIf(not TQDM_INSTALLED, "optional tqdm module is not installed")