High,
we put a beta relase of the new NeXus API supporting HDF-5
with the following added features into the NeXus repository:
- F77 API
- a couple of bug fixes
- NXdict has been adapted. You now can specify a -chunk {dim1, dim2, ..}
option for SDS which determines the chunksize.
- The Java API has been adapted. It has been tested under True64Unix only
so far.
We will start writing data HDF-5 data files soon using the new API.
Writing compressed data sets in slabs works only reasonably fast with
the newest HDF-5 version 1.4.2patchlevel1. With any other version it
takes 10-15 min on an alpha to write a 16MB chunk of compressed data!
The HDF-5 performance for the above mentioned operation is also dependent
on the HDF-5 cache size. If the cache size is set to a value
correponding to the slab size another factor of 4-5 can be gained in
performance. Unfortunately this cache size can only be set at file
opening time. In order not to break to much existing code we introduced
a new static variable and a new API function:
NX_status NXsetcache(long newSize)
for setting the cache size. This function must be called immediatetly
BEFORE NXopen. This is only of interest to NeXus users who write large
data sets, anybody else is well served with the default 1MB of cache.
The HDF-5 team, i.e. Elena Pourmel, is working at these performance
issues.
In the pipeline: A new NeXus-API for IDL which uses a shared library much
like the Java API. This would then support both HDF-4 and HDF-5 in IDL.
This is mostly finished but we need to do some more testing.
Regards
Uwe Filges
Mark Koennecke