> Hi Nbody,
> in my view a fundamental parameter of Nbody data formats
> is IO speed. This becomes more essential as CPUs learn to handle
> larger numbers of particles. Binary formats are therefore
> better than ascii in that respect. We need tests comparing
> IO speed for FITS tables and raw binary files, then we
> can decide. Hierarchical data structures are probably too
> complex as a data exchange format.
>
> Marc Balcells
> balcells@astro.rug.nl
I think the choice depends on the primary purpose of the "universal"
N-body data format. When I'm running my code, I use binary format for most
I/O operations, for speed, space saving, and floating point precision.
This allows me to record data at frequent intervals for later analysis.
But any data used for initial conditions, say, need not be in a compact
format, since the cost of accessing the data is only incurred once. In
fact, I'd suggest it would be better that this data be in as
human-readable a form as possible, both for compatibility between
platforms and for easy identification. I don't envision myself storing
data in this format at frequent intervals; rather I'd have a small program
to convert my specialized binary output to one more suitable for sharing
between other researchers.
Cheers,
Derek
___ ___ ____ ------------------------------------------------------
____ __ _____ | Derek C. Richardson | Tel: (416) 978-6879/8494 |
_____ __ ____ | CITA, McLennan Labs | (416) 487-7014 (home) |
____ __ __ __ | U of T, Toronto, Ont. | Fax: (416) 978-3921 |
___ ___ __ _ | CANADA M5S 1A7 | richards@cita.utoronto.ca |
------------------------------------------------------