Pure XML model for VOTable
roy at cacr.caltech.edu
Tue Apr 29 07:56:52 PDT 2003
> It seems we are trying to make XML efficient and therby not ending up
> with xml anymore. XML is efficient - it can be greatly compressed
> and we may use DIME to attach binary objects.
(1) The original idea was to use XML to describe the metadata and then use
other formats -- binary, FITS, XML -- for the bulk data. Therefore in the
current VOTable, the metadata section is actually a schema -- it says that
every record of the data section has the given form.
(2) The DIME mechanism is like MIME -- it just says "this is a table", and
doesn't specify or connect to the complex metadata of the table that is
presumed to be somewhere else. I don't see how DIME is particularly better
than the VOTable binary or FITS binary that we already have.
(3) I understand your point that using XML *only* -- for data and
metadata -- makes much of the software easier, because we can use the
sophisticated XML tools that are now quite mature. I understand your point
that compression can reduce the volume of XML files. However I would be very
interested in timing of a billion records of compressed XML versus a billion
records in FITS and raw binary.
(4) In the pure XML model that you support, are you suggesting that the
metadata section be real XML schema? Something like this?
<xsd:element name="FruitName" type="xsd:string"/>
<xsd:element name="Colour" type="xsd:string"/>
(5) How do we manage subclassing? How do we have a generic table schema that
is agreed by the IVOA committees, yet also have this record schema that is
different for each table? How do UCD and units fit into this?
(6) I look forward to your detailed presentation on an alternative to the
current VOTable. I really hope there be time in Cambridge for this!
Your open-minded friend
More information about the votable