[NeXus-committee] More DECTRIS

"V. Armando Solé" sole at esrf.fr
Mon Jan 14 16:09:30 GMT 2013


On 14/01/2013 16:52, Mark Koennecke wrote:
> Hi,
>
> Let us assume that NP is the number of frames.
>
> Then the optimal solution in terms of NeXus (and may be HDF-5) would 
> be to have an
> array data[NP,xsize,ysize]. The problem is that we cannot have the 
> optimum because
> HDF-5 cannot compress in parallel and DECTRIS does not get the write 
> performance
> they want without compression and accepting the limitation that they 
> get to define the
> computer delivered with the detector and are in no position to force 
> parallel file systems
> on their users. So we need to settle for second best.
>
> This leaves us with the option to put all the links into one NXdata or 
> have a NXdata per chunk.

Since [NP, image_row_size, image_column_size] (not xsize, ysize because 
it is just the transpose of the image in HDF5 default C notation)  is 
the "desired" final result and that would fall in a single NXdata group, 
all the links should be in a single NXdata group.

We should leave DECTRIS do their things and cook ourselves an attribute 
to the NXdata group containing the links that tells us what to do with 
the links.

Armando



More information about the NeXus-committee mailing list