Keep in mind that dx -memory n, only give n MB available to dx.  32 MB
would be barely enough to get the exec going.  In addition, the exec uses
32-bit addressing.  So, you can only give 2 GB to the exec, unless Greg has
changed this in a more recent release.

I use much larger meshes from netCDFs without problem on AIX, Linux and
Windows.  I typically read one time step at time.  The data are on
irregular positions and regular connections (quads or cubes).  But since
the mesh is constant, I read that in once and cache it.  Subsequent reads
replace the data component, which is more efficient.  Granted, the size of
your data should easily fit in memory.  NetCDF certainly doesn't do
anything to be conservative of storage except having a simple notion of
what DX calls product arrays.






Nils Smeds <[EMAIL PROTECTED]>@opendx.watson.ibm.com on 07/16/2001 10:34:38
PM

Please respond to [email protected]

Sent by:  [EMAIL PROTECTED]


To:   [email protected]
cc:
Subject:  Re: [opendx-dev] netCDF import of time series data



> By default a series is loaded into memory all at once so you could be
> definitely outstripping available resources once you enlarge your
> series.
>

I am aware of that, but this is a time series of 6 time steps where each
time step is a scalar field on 64x32 points. The positions are the same
for all fields and connected through the same quad connection
throughout. This is a total of ~200k of data in the raw netCDF file.
Even though it blows up inside the visualizer due to the extra
information you need to keep available it still seems odd that it does
not work on a system with 6GB of RAM.

In fact, I now repeated the test using a 6x6 mesh instead and the import
module crashes if I have 6 time steps, but works ok when I have 5 time
steps. Admittedly, this was on a 128 MB linux system and using the
command line:

     dx -memory 32 -local -host localhost &

But this same system and command line happily displays an 8x8x8 3D mesh
and the 64x32 mesh provided the time series is no longer than 5 time
steps.

I am by no means ruling out that I am doing something wrong. Possibly in
the structure of my netCDF file. But I don't believe it is a system
resource issue.

Anyway, thanks for the input. I'd be more than thankful if anyone could
find out where the problem lies.

Cheers,

/Nils
--
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   Nils Smeds
   HPCSU / Sci & Techn.              e-mail: [EMAIL PROTECTED]
   UNSW, NSW 2052                    Voice:  +61-2-9385 6915
   Sydney                            Fax:    +61-2-9385 ----
   AUSTRALIA                         Office: Red Centre, West, Rm 2075
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   Nils Smeds                        http://www.pdc.kth.se/
   Center for Parallel Computers     e-mail: [EMAIL PROTECTED]
   Royal Institute of Technology     Voice:  +46-8-7909115
   KTH                               Fax:    +46-8-247784
   S-100 44 Stockholm, Sweden        Office: OB2, room 1546
-----------------------------------------------------------------------



Reply via email to