[netCDF #GDB-514857]: RE: using netCDF for storing data from realtime source

Hi Sjur,

After replying, I just thought of one more possibility that might explain why 
you were seeing increasing times as you added more records.

If you were also changing the schema of the netCDF file, by adding new 
attributes, dimensions, or variables as you added new records, that would 
definitely cause a slowdown.  There is a fixed amount of space for the file 
schema and attributes at the beginning of the file.  When more space is 
required for the file schema (also called the "file header"), the data is 
recopied after more space is allocated.

This is easy to identify in the C interface, where you have to re-enter "define 
mode" to change the schema, but the C++ interface tries to hide "define mode" 
from the user (probably a mistake) by making the calls for you to enter and 
leave define mode when adding new variables, dimension, or attributes, renaming 
anything with longer names, or extending the length of an attribute.  So just 
putting a longer string value in an attribute might be enough to make the I/O 
appear to slow down, since the data would be recopied when you extended the 
attribute value to require more space in the file header.  Is it possible you 
are doing something like that when you add records, making it appear as if 
writing the records slows down, but actually the time is taken by an implicit 
call to nc_enddef() in the C interface to rewrite the file header and all the 
data?

--Russ

Russ Rew                                         UCAR Unidata Program
address@hidden                     http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: GDB-514857
Department: Support netCDF
Priority: Normal
Status: Closed