[svn-r22031] Purpose:

  Task for HDFFV-7862 - Select data by chunk direction to improve performance in h5repack

Description:
  h5repack sometimes became very slow when handling big chunked datasets in
  certain cases. (when chunk boundary doesn't match with a hyperslab boundary.)
  The main issue was from figuring out a hypeslab without considering chunk
  boundary to read from and write to such datasets.
  The update was made to figure out a better hyperslab unit with considering
  chunk boundary to improve performance for such cases prior to the update.


Tested:
  jam (linux32-LE), koala (linux64-LE), ostrich (linuxppc64-BE), tejeda (mac32-LE), linew (solaris-BE), Windows

1 files changed, 2 insertions, 2 deletions

diff --git a/tools/lib/h5tools_utils.c b/tools/lib/h5tools_utils.c
index 4505609..11ab5f3 100644
--- a/tools/lib/h5tools_utils.c
+++ b/tools/lib/h5tools_utils.c
@@ -49,9 +49,9 @@ static const char  *h5tools_progname = "h5tools";
  * largest value suitable for your machine (for testing use a small value).
  */
 /* Maximum size used in a call to malloc for a dataset */
-hsize_t H5TOOLS_MALLOCSIZE = (128 * 1024 * 1024);
+hsize_t H5TOOLS_MALLOCSIZE = (256 * 1024 * 1024);  /* 256 MB */
 /* size of hyperslab buffer when a dataset is bigger than H5TOOLS_MALLOCSIZE */
-hsize_t H5TOOLS_BUFSIZE = (1024 * 1024);
+hsize_t H5TOOLS_BUFSIZE = ( 32 * 1024 * 1024);  /* 32 MB */
 
 
 /* ``parallel_print'' variables */