gzipped images in SIAP 1.0 (fwd)
Rob Seaman
seaman at noao.edu
Wed May 30 12:15:45 PDT 2007
Steve Allen says:
> I would tread carefully until it is clear whether and when the FITS
> community at large is prepared for that sort of usage.
The bigger question is whether the FITS community is ready for a
registry of conventions. Treading carefully is as much a watchword
for FITS as for Col. Gadsden (http://en.wikipedia.org/wiki/
Image:Gadsden_flag.svg). (I'm writing this from his grandson's
purchase.)
Tile compression is such a useful thing that one has hope it will
drive application/fits to similar levels of utility. I still think
it arguable that a tile compressed file be represented as image/fits
- but by the time this is resolved, the meaning of application/fits
will be similarly enhanced.
As usual, Doug clarifies the underlying issues:
> I think we need to keep two issues clearly separated here: 1) what
> the client application wants to see, and 2) what happens at the
> level of the HTTP protocol.
This is an admirable goal. I believe it is challenged on two
fronts. The first is that mime-like semantics are insufficient to
keep these separate. For instance, what does:
Content-Type: application/x-gzip
Content-Encoding: gzip
(or Transfer-Encoding) require of the client? I won't further
belabor the discussions to date.
The second issue is that efficient throughput is a function of the
entire system, not just a separable transport layer, HTTP or
otherwise. Ideally services should produce data benefiting from
native compression such as FITS Rice tile compression. This
compression could be applied with the first copy of the data, and
could remain in effect all the way to the client. Even a client that
requests gzip'ed formats must eventually uncompress the data to
access header keywords as well as pixels. A client that requests
tile compressed formats may frequently never need to uncompress those
data. In those cases in which on-the-fly uncompression is needed,
this is a much lighter weight function for tiled compression than for
whole-file gzip.
Rob
More information about the dal
mailing list