[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

20020402: NetCDF on a Fujitsu VPP5000

Dear Wolfgang,

> To: address@hidden,
> To: Frank Schmitz <address@hidden>
> From: Wolfgang Kouker <address@hidden>
> Subject: [Fwd: away from my email]
> Organization: Forschungszentrum Karlsruhe GmbH
> Keywords: 200203251426.g2PEQga06404

The above message contained the following:

> Dear Colleagues,
> 
> We have promoted using netcdf as a standard dataset when exchanging data
> among working groups. Since netcdf allows a wide vriety for structuring
> data, I am made responsible for setting up a standard interface for
> gridded data. The interface including a small test driver is provided in
> the attachment (netcdf.sh). The output netcdf.nc file is attached, too.
> 
> The test seems to work well on some workstations like SGI and DEC Alpha.
> The installation of netcdf (3.5.0) was also straight foreward on these
> machines.
> 
> Our standard number chruncher, however, is a Fujitsu VPP 5000, operated
> by rank Schmitz. There, the installation of netcdf was possible only
> with a workaround provided with Steve's support. Even though the
> compilation was successfull now, the interface doesn't work. Please let
> me outline the problems as follows:
> 
>    * The test "stops" somewhere in nf90_get_var (called in cdfread),
>      i.e. it simply doesn't do anythin visble. It does especially not
>      give control back to OS!
>    * Editing the test in the main program by activationg the "stop"
>      immediately before opening the netcdf datet the 2nd time, a file is
>      created (netcdf.vpp). Dumping (ncdump) the dataset leads to
>      netcdf.vppdump on VPP. Dumping it on the alpha leads to a floating
>      exception. The last lines of the dump are given below. The dump
>      looks good for all the attribute and dimension information but the
>      data is as named: garbage...
>    * Dumping the netcdf file created on the alpha on the vpp leads to a
>      good result, thus I believe there is something wrong writing the
>      data.
> 
> I will be out of office from now until middle of April, but I will
> appreciate any assistence making netcdf available on our vpp.
> 
> best wishes, Wolfgang
> 
> --
> Dr. Wolfgang Kouker
> Inst. f. Meteorologie und Klimaforschung
> Forschungszentrum Karlsruhe
> Hermann-von-Helmholtz-Platz 1
> D-76344 Eggenstein-Leopoldshafen, Germany
> Tel. (++49)7247-824036, Fax (++49)7247-824742
> http://www.fzk.de/imk/imk2/kasima

Unfortunately, we do not have access to a Fujitsu VPP5000; consequently,
it is very difficult for us to diagnose problems with the netCDF
package on that system. Also, we don't have the resources necessary for
searching foreign code for usage errors.

If the "make test" portion of the netCDF installation procedure
completed successfully, then the netCDF package should be correct. The
tests are very extensive. Did the "make test" complete successfully?
If not, then would you please send me the standard output and standard
error from executing the "make test" command.

Alternatively, if you could get us an account on the Fujitsu VPP5000,
then we might be able to diagnose the problem.

Regards,
Steve Emmerson   <http://www.unidata.ucar.edu>