table metadata and the registry

[Paul Harrison]

The only metadata that we want in the registry is the *necessary*  
metadata to interact with the functional interface of the service -  
if the functional interface of the service requires knowledge of the  
column names/ucds then we think that metadata should be cached in the  
registry.

This whole argument concerns the efficiency and response times of a  
grid of such services to the VO end user. We want  cached metadata to  
speed up the selection of appropriate services for particular  
queries. It is all about having a "grid view" rather than a "point  
service view" - I thought that this ability to "combine data" was one  
of the central promises of the VO.

As I said before - if the placing of table/column metadata in the  
registry is deprecated  by the TAP group, then they need to  
simultaneously come up with a design/specification/implementation of  
a "table metadata  registry" - otherwise the VO slows down for the  
end user. This metadata does not *have* to go in the existing  
registry, but it sure is the quickest route to to the required  
functionality.

Paul Harrison
ESO Garching
www.eso.org

On 10.05.2007, at 03:33, Roy Williams wrote:

>
> Tony Linde wrote:
>> I don't agree, Ray. The table/column information is as much a part  
>> of the
>> description of the service as any of the aother metadata.  
>> Splitting it out
>> is arbitrary: why is the table/column information different from  
>> the rest of
>> the metadata?
>>
> The metadata for a book on Amazon.com includes images of the front  
> and back cover, and full size images of the inside of the book,  
> lists of Statistically Improbable Phrases from the book, and lists  
> of who wants to sell secondhand copies. If we are expanding out the  
> concept of metadata, why stop with table/column information? How  
> about we include the full text of the journal paper that describes  
> the dataset? Perhaps a few hundred sample records from the dataset?  
> Comments from users, IDL scripts that people have used to analyze  
> the data? In fact, why not go the whole way and just copy the whole  
> thing into the registry. All data is really metadata. Centralize  
> copy and replicate.
>
> Sorry, getting a bit silly here.......
> :-)
> Roy
>