RE: next steps / open issues in DASL framework

The question of arbiter discovery is an open question, 
I think. I was thinking it would get addressed on the
second release of SEARCH, when cross-repository searching
was addressed. I don't think that arbitrary resources
(e.g., a document) need to identify possible arbiters.
However, whether the collection itself should do that is
a different matter. Since the question of how to find
arbiters is open, I'm not sure that the collections should
be _required_ to provide that, but it seems like a very
useful thing to do if there is a reasonable way to do it.
So, perhaps collections should at least be allowed to
do that.

How would a collection provide information on arbiters? 
Would there be a well known property on the collection 
itself giving the URL of an arbiter?

Alan

-----Original Message-----
From: Elias Sinderson [mailto:elias@cse.ucsc.edu]
Sent: Thursday, March 14, 2002 1:35 PM
To: dasl
Subject: Re: next steps / open issues in DASL framework


"Babich, Alan" wrote:

> (1) Why couldn't there be multiple arbiters that searched a collection?
That
> would seem to be required if and when we transition to arbiters that can
> search across multiple collections. (The original arbiter would still work
> for upward compatibility, and one or more new multiple-collection arbiters
> would come into existence.)

We're in agreement on this - I would object to anything in the spec. that
implied otherwise.

> (2) Therefore, it seems that requiring collections be aware of all the
> arbiters that can search them is not acceptable. Why should collections
care
> what arbiters can search them anyway?

The collections may not care, however, the open question is one of arbiter
discovery - How can I find out what arbiter to direct my query to if I want
to
search through collection X. I think that interrogating the collection
directly
is the most direct way to address this.

> (3) Why should a collection be forced to act as a arbiter? That would be
an
> undue burden and bad layering.

I agree, resources in general should not be forced to act as an arbiter, but
should be able to direct someone to a search arbiter, if one exists, that
can
be used on that resource. This does not imply that a resource should know
about
every arbiter, a generally impossible requirement to satisfy, but a minimum
requirement should be that a resource should identify a default arbiter that
can be used.


Elias

Received on Thursday, 14 March 2002 17:44:55 UTC