segmentation fault using multiple netcf files and xdfopen

Arlindo da Silva dasilva at ALUM.MIT.EDU
Thu Feb 7 09:05:07 EST 2008


On Feb 7, 2008 12:15 AM, Adam Sobel <ahs129 at columbia.edu> wrote:

> I have a set of netcdf files, each is 1 day.  I have a .ctl file so I can
> view a bunch of them together in sequence using xdfopen.  The ctl file
> is below.



> If I set t to the whole time range (in this case set t 1 12) I can view
> the variables u or v just fine, e.g. make animations for the whole
> period, looks good.  But then if I do
>
> define rvort=hcurl(u,v)
>
> I get a segmentation fault


This is a well known problem in the v1.8/v1.9 series (but not v1.7b9 and new
v2). There is a memory leak in the sdfopen/xdfopen part of the code that
causes the amount of memory to increase with the size of the timeseries
until you ran out of memory. (On a linux system you can watch this happen by
opening another window and executing "top"). I usually experience this
problem with much longer data sets, but all depends on your system load and
horizontal resolution.

I have a patched a version of GrADS for use at NASA/GMAO which is built with
the sdfopen/xdfopen code from  v1.7b9 sdf/xdfopen code, with some of the
bugfixes up to v1.9b4. Since it *may* not address all NetCDF files supported
by v1.9b4, COLA recommended that this patch not be included in v1.9.0-rc1.
Their recommended action is to prepare a ctl file with "DTYPE netcdf"  and
use the "open" command. Another option is the just released v2, but this
version is still in active development and is not as stable as v1.9 yet.
BTW, this problem is common to netcdf and hdf files alike.

     Arlindo


--
Arlindo da Silva
dasilva at alum.mit.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://gradsusr.org/pipermail/gradsusr/attachments/20080207/fbd21434/attachment.html 


More information about the gradsusr mailing list