Richard Gerber (
Thu, 17 Nov 1994 14:09:23 -0800 (PST)

> just noted your URL; since you're potentially interested in this,
> check out the NEMO url, but in particular the Data Interchange Format
> issues
> In the 'table of contents' you should an entry to Nbody Data Interchange
> Format. You might be interested in getting on the e-mail exploder.
> I would also like to exchange some sample data with you, either SPH or
> pure Nbody, of e.g. an interesting simulation from one of your papers,
> or for other interests. I'm using them not only as archive but as
> sample datasets that people can try their converters on. We've been
> finding all kinds of interesting problems this way.

Hi Peter,

Yes, please do add my name to the list.
I very much like your nemo page and I've downloaded some
files to take a look at. I'll get you some data and some
simple routines in Fortran and/or C, whichever you
prefer, to read it.

As for the data interchange format, I have these initial
thoughts. You can forward them to the list if it
seems appropriate. It doesn't seem like there's much
discussion going on yet.


I've been using the NCSA HDF library routines to store
data and I've been extremely pleased with the results. The
reasons are many:

1. Transparent file transport between platforms. I have
used the same data files on Crays, Suns, SGIs, HPs, Macs,
and my Amiga (for which I did a port of HDF) with
no problems. So far I've been unable to compile
the library on the Connection Machine at NASA Ames, however.
2. Ease of coding in both FORTRAN and C. A couple of calls
and that's it.
3. Data compression built in.
4. HDF is self-describing. That is, there are calls to retreive
array dimensions, data types, etc.
5. HDF includes an image format. You can mix data and images in
the same file.
6. Speed. Since it's a binary format, I call HDF directly from
my main programs and have one less conversion step in going from
machine output -> visualization/analysis.

Of course it's not perfect. HDF is really set up to deal with
rectangularly gridded data. The HDF Vset extension, which is supposed
to be much more flexible, has always given me problems.

Currently, I place the following vectors into a 10xN 2D array and
then store that.

"color" (some ID tag),
density (for SPH), and
internal energy or sound speed (for SPH).

Richard Gerber