951017: netCDF question

Jim,

> Subject: netCDF question
> To:       address@hidden
> From: address@hidden (Jim Hines (awdnsun) 472-6708)
> Organization: .
> Keywords: 199510171555.AA05908

In the above message you wrote:
 
> I just started playing with netCDF and do not understand something.
> i wrote 2 netCDF
> 
> exp1
> **************************************
> netcdf 3200proto {
> dimensions:
>       time = UNLIMITED ;
> variables:
>       short EVAP(time) ;
>       short EVAPQ(time);
> }
> 
> exp2
> **************************************
> netcdf 3200proto {
> dimensions:
>       time = UNLIMITED ;
> variables:
>       short EVAP(time);
> }
> ***************************************
> 
> 
> I then used ncgen (with -b option) to build the the netCDF file
> 
> the size of the netCDF file of exp1 was 120 bytes long
> the size of the netCDF file of exp2 was 80 bytes long
> everything ok
> 
> 
> I then wrote the following fortran program and ran it on both netCDF file
> I think it added 100 values to the netCDF file
> *************************************************************
> c
>       INCLUDE 'netcdf.inc'
>       integer ncid,rcode
>       integer evapid
>       integer*2 ival
> 
>       integer index(1)
>  
>       ncid=ncopn('coop.nc',NCWRITE,rcode)
>       write(*,*)"ncid=",ncid,"   rcode=",rcode
> 
>      
> c  get variable ID's
>       evapid=ncvid(ncid,'EVAP',rcode)
>       write(*,*)"evap id=",evapid,"    rcode=",rcode
> 
>       do 10 i=1,100
>         index(1)=i
>         ival=i
>         call ncvpt1 (ncid,evapid,index,ival,rcode)
>  10     write(*,*)"ncvpt1    rcode=",rcode,i
> 
> 
> c  must use ncclos or ncdump does not work properly
> 
>       call ncclos(ncid,rcode)
>       write(*,*)"rcode close =",rcode
> 
>       stop
>       end
> ************************************************************
> the program seamed to work but
> the size of the netCDF file in exp1 was now 920 bytes
> the size of the netCDF file in exp2 was now 282 bytes
> 
> the size of the exp2 example seamed to be about right..
> the increase in space of exp1 is about double of what
> i would have expected
> 
> any ideas on what i might be doing wrong

You're not doing anything wrong.  The discrepancy between the actual
sizes and the "expected" ones results from using a too-simple model of
the netCDF file format.  In the case with two short variables in each
record, the netCDF library actually writes two 4-byte variables per
record (the 2-byte shorts are expanded to 4-bytes by the XDR library).
Consequently, the size of the data is 2*4*100 = 800 bytes.  In the
case with a single short variable in each record, the netCDF library
is clever enough to optimize the I/O and write the entire array using
2-byte variables rather than 4-byte ones.  Consequently, the size of the
data is just 2*100 = 200 bytes.

--------
Steve Emmerson   <address@hidden>