Howdy,
I started trying out netcdf 4.1.1-rc1 (with hdf5-1.8.4-patch1), and
noticed that the write times from my application went up by a factor
of 1.8 to 2 compared to the 20091122 snapshot. The problem showed up
for larger files: say, 25MB of data (compressed) per record (in this
case, about 40 3d arrays per time level with dimensions of
119x119x120). I did a test with a smaller data per record, and rc1
doesn't show any slowdown (perhaps even a little faster than 20091122).
I discovered that the problem is the default chunksizes: Ncdump -hs
shows that the file created with 20091122 has _ChunkSizes = 1, 120,
119, 119 (the same as the actual dimensions), but rc1 is setting
_ChunkSizes = 1, 102, 101, 101, which turns out to be much less
efficient. When I altered my code to specify the chunksize to be the
variable dimensions, the write time for rc1 went back down, too.
Any reason for the change in how the default chunksizes are set? Is
there some arbitrary maximum default chunk dimension that got reduced?
Best,
-- Ted
__________________________________________________________
| Edward Mansell <ted.mansell@xxxxxxxx>
| National Severe Storms Laboratory
| 120 David L. Boren Blvd.
| Room 4354
| Norman, OK 73072
| Phone: (405) 325-6177 http://www.cimms.ou.edu/~mansell
| FAX: (405) 325-2316
|
|
----------------------------------------------------------------------------
|
| "The contents of this message are mine personally and
| do not reflect any position of the U.S. Government or NOAA."
|
|
----------------------------------------------------------------------------
