NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
Dan Kokron a écrit :
Hello all, I have an application that can use either NetCDF-3.5.0 or HDF to write files. Both libraries give about the same performance, but the routines where each library spends it's time are very different. HDF spends most of it's time writing while NetCDF spends most of it's timecreating files. Specifically, NetCDF spends time in these calls ...(32 seconds for 21 calls) call ncvpt (fid, timeid, corner, edges, 0, rc) (104 seconds for 168 calls) call ncendf ( fid, rc ) I'm using a build of netCDF that someone else compiled. I'm not sure what compile flags he used. Has anyone else seen this behavior before?
When creating files with netcdf if the flag NC_FILL (see function nc_set_fill) is set each time a nc_enddef call is issued all new var data is written with the _FillValue (specified or implicit). If you are sure to write all data you may set the flag NC_NOFILL gaining many time since data is written only once. On an other hand a way to gain more time (if it is possible) is to create the file and define all dim/var/attributes at the beginning: only one nc_enddef call (instead of 168 as it seams according to your output). Defining a variable writing its value, defining another var, writing it's value(...) is very time consuming since the var values are to be moved each time because of the increasing size of the file header (which contains the definitions) and maybe the data values also if new variables belongs to the unlimited dimension. If you cannot define everything on file creation you may have a look at the nc__enddef (double underscore '_') in order to allocate enough space in header and avoid its growth. Expecting it can help, -- Ph. Poilbarbe CLS, Space Oceanography Division -- Philippe Poilbarbe CLS Océanographie Spatiale mailto:Philippe.Poilbarbe@xxxxxx phoneto:+33(5)61394727 Parc technologique du canal 8-10, Rue Hermes 31520 Ramonville St-Agne ============================================================================== To unsubscribe netcdfgroup, visit: http://www.unidata.ucar.edu/mailing-list-delete-form.html ==============================================================================
netcdfgroup archives: