VOSpace home page

VO and VOSpace data formats

This is a discussion page looking at defining and registering a list of standard VO data formats.

Although the VOSpace team will probably want to register VOStandard resources for the VO data formats, this discussion is not intended to be about using registry URIs vs MIME types to describe data formats.

All I am suggesting at the moment is that we work together to produce a list of the core VO data formats. As far as I know, at the moment we don't have a list of the common data formats used within the VO. If we do already have one, then please let me know.

Initially, this can be just a wiki page that describes the formats, provides links to the relevant RFC documents, lists the common MIME types associated with each format, specifies the VO standard MIME type(s) we want to use and describes how the formats are used within astronomy and the VO.

The information collected on the wiki may form the basis for a VOStandard resource that registers the common data formats, but the initial step is to collect the information.

-- DaveMorris - 29 May 2007


Concepts

The recent discussion thread about data formats and MIME types on the DAL mailing list started with a question about "gzipped images in SIAP 1.0" and has openned up into a wider discussion about MIME types and HTTP headers.

The thread has highlighted the fact that there are at least three different concepts that we need to represent when transferring data.

  • What it contains : tabular data, an image etc
  • How it is represented : the primary data format, FITS, VOTable etc (plus modifiers e.g. binary VOTable and tiled FITS)
  • How it is transferred : e.g. zip, tar, gzip, targzip compressed files or streams

The discussion has looked at various options for representing these concepts in the HTTP header fields :

  • Content-type
  • Content-encoding
  • Content-disposition

We have been struggling to solve similar problems within the VOSpace group. The problems we are trying to solve within VOSpace may be slightly more complex for a number of reasons :

  • VOSpace supports asynchronous 3rd party transfers, where the client arranges for data to be sent from one VOSpace service to another. This means that the data transfer occurs outside the scope of the SOAP or REST service call, and so the client cannot use the HTTP request fields to indicate what format it would like the data to be transferred in.
  • VOSpace needs to be able to support other data transfer protocols, like Ftp and GridFtp, which may not have the equivalent metadata fields for Content-type and Content-encoding etc.
  • VOSpace needs to be able to distinguish between
    • "I am going to send you a zip file containing a set of FITS images, and I want you to store it as a zipfile"
    • "I am going to send you a zip file containing a set of FITS images, and I want you to unpack it and store|process the individual FITS images"

However, the core problems of deciding which formats to use, and how to refer to them are the same.

The current VOSpace specification introduces the concept of registered URIs to refer to 'views' which unfortunately attempts to represent at least two of the concepts what it contains and how it is represented using a single URI. We know this is not sufficient for what we need, and plan to revisit this fairly soon.

Using registered URIs to identify content and formats has proved to be useful, certainly within the context of VOSpace. However, I appreciate that other groups may prefer to use MIME types or something similar, particularly for HTTP based access.

I think we should work together with the other groups to produce a list of the main VO data formats, their corresponding MIME types and see if we can identify an inheritance hierarchy where it is appropriate.

For VOSpace services, we would probably want to register these formats in a VOStandard registry resource, enabling us to use registry URIs to refer to them.

Even if we eventually decide not to use the registered URIs in services and applications, creating a list of the main VO data formats and their corresponding MIME types will provide a useful resource for developers working on VO projects.

The first step is to make an initial list of the formats, define the corresponding MIME types and the inheritance hierarchy.


Data formats and container formats

My initial guess is that we need two types of format, a data format and a container format.

  • A data format is a file format that contains a representation of the data, e.g. FITS or VOTable
  • A container format is a file format that contains other files, e.g. zip or tar

Ideally, we would want to be able to come up with a standard vocabulary that could represent the following concepts :

  • The selected images are available as individual FITS/image files.
  • The selected images are available as a zip file containing the individual FITS/image files

In this model FITS may be a special case. From my admittedly basic understanding of FITS, it is possible for FITS to be both a data format (a FITS file containing tabular or image data) and a container format (one FITS file containing multiple images or tables).

I defer to others who have a much better understanding of the FITS format and its useage within astronomy to define this in more detail.

Ideally, we would want to be able to come up with a standard vocabulary that could represent the following concepts :

  • The selected images are available as individual FITS/image files.
  • The selected images are available as a single FITS file containing multiple images

I don't know if we can define a simple MIME type that distinguishes between these two.

At the moment I'm looking at building a common list of what formats we want to be able to represent, and our initial best-guess at how we want to describe them. If this particular case is overly complex and rarely used, then we label it as 'there be dragons' and move on.

Specialization and inheritance

Some data formats may be specializations of existing formats, and may inherit the MIME type or other details from their parent format.

Java JAR format

An example of this is the Java Archive format. From the JAR file specification :

  • A JAR file is essentially a ZIP file that contains an optional META-INF directory ... [that contains] specific files and directories .... that are recognized and interpreted by the Java 2 Platform.

This implies that the JAR format extends the ZIP specification, with additional metadata in the META-INF directory. As far as I know, there isn't a specific MIME-type defined for Java JAR files, so the JAR format would 'inherit' the MIME type string from the ZIP format.

This particular example may not be useful for astronomers, but I listed it here because it is an already established extension format that developers will be familiar with.

VOSpace archive format

What I think will be useful for astronomers is to be able to say

  • Archive this branch of my VOSpace as a zip or gzip file and save it to backup space
then at a later date
  • Load this backup into VOSpace and restore all the VOSpace metadata
in order to support this, we may need to define a similar extension to the zip or gzip format that includes VOSpace metadata.

Data set specific FITS format

Having talked about this with some of our astronomers, one thing they suggested would be useful is be to be able to define types that represent data from specific data sets.

If we had a VO data type that represented FITS image, then they would like to be able to define a new type that represented a FITS image from a specific data set.

This new type would extend the definition of FITS image, inheriting all of the properties and behaviour defined fo a generic FITS image, and add details of the specific FITS header fields that that particular data set used in their files.

Defining this as an extension type means that tools and applications that don't understand the specific type could handle the data as generic FITS, but tools that did understand the more specific type would be able to make use of the additional metadata that described what the data set specific header fields meant.

At this point I don't know if this is a GoodIdea or not, or whether this information should be encoded in the content type or in a separate property. However, our astronomers seemed to think that the ability to distinguish between a generic FITS file and a FITS file from a specific data set was important.

How we enable them to make this distinction is up for discussion.


Edit | Attach | Watch | Print version | History: r7 < r6 < r5 < r4 < r3 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r5 - 2007-05-29 - DaveMorris
 
This site is powered by the TWiki collaboration platform Powered by Perl This site is powered by the TWiki collaboration platformCopyright © 2008-2024 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback