C[omp]ute

From: "andrew cooke" <andrew@...>

Date: Thu, 8 Dec 2005 10:42:40 -0300 (CLST)

This is an email from a list I'm subscribed to that discusses the "Virtual
Observatory" - astronomy across the 'net.  They have been using URIs of
the form ivo://noao.edu/nsa/some/object, but this email is a very clear,
readable argument for the use of http://... instead.

Andrew

---------------------------- Original Message ----------------------------
Subject: Re: Multi-conference report: VO and SW
From:    "Norman Gray"
Date:    Thu, December 8, 2005 10:15
To:      "Roy Williams"
Cc:      semantics@...
--------------------------------------------------------------------------

Roy,

How's it going?

On 2005 Dec 4 , at 22.58, roy wrote:

> I would appreciate an elaboration of why it ivo: is bad.

As Tony says, I'm to some extent the messenger here.  Myself, I'm  
probably more persuaded than not that URIs are best when they are  
retrievable, but I'm diffident about saying so too loudly, because I   do
agree that things like ivo: seem to provide the strongest
counterexamples.  However, I'll make the case as strongly as I can   here.
 And honestly, I _have_ tried to keep this short!



The Web Architecture document seems broadly agreed to be a practical  
rather than theological document, and a description of what in
retrospect made the web `work'.  It's from that hindsight point of   view
that HTTP was identified as having the following almost
magically good properties:

   1. everyone with a web server can mint identifiers (ie, it's non-
exclusive and hugely extensible)

   2. HTTP URLs have all the properties you want in an identifier,
plus retrievability

   3. HTTP GET has very clean and intelligible semantics, which are
simple to implement and integrate in other applications (ie, you   don't
need a resolver)

Now, the VO isn't interested in property (1) (for better or for
worse), so we can skip that.

(2) is essentially the argument that if you want something other than   a
URL for your URI, then you're going to have to make a positive case   why
direct retrievability is a practical disadvantage.

The big difference between URIs and URLs is, as we all know, that the  
former are for labelling concepts, and aren't guaranteed to be
dereferencable, much less immediately retrievable.  However a long- 
established bit of good practice here is to make your URIs
retrievable and to provide _something_ useful at the end of it, be it   a
human-readable description of your namespace, or of your ontology,   or
whatever you personally mean by the concept thus labelled.

This argument is therefore flatly contra your:

> Another example is the XML namespace. This is a silly name for it:
http://www.w3.org/2001/XMLSchema-instance. I would suggest that
> xmlns://www.w3.org/2001/XMLSchema-instance would be better.

The two are precisely equivalent as far as the machines are
concerned, but the latter has given up the free opportunities for  
documentation or clarification that the former offers, for no real  
benefit.    To get those advantages back, you'd have to expensively  
define the xmlns: scheme, and specify and maintain resolvers for it.

As well as the gratis practicality, the recommendation that URIs   should
be URLs undermines any claim that there is still a useful high-  level
distinction between URIs and URLs, beyond the incidental one   that some
are retrievable and some are not (this goes as far back as   2001
<http://www.w3.org/TR/uri-clarification/> with other remarks in  
<http://www.ietf.org/rfc/rfc3305.txt>).

So the response to your objection

> I really dislike this practice of making non-URLs look like URLs,   it is
> confusing.

is: `so make them into URLs'.  You're going to have to document your  
`non-URL' somewhere, so it might as well be there.  `But the URI/URL  
specifies a resource, not a namespace', you say: so let the URL
retrieve the resource.  Why make things hard by adding a resolver?

The answer is, of course, because we want the IVOA Registry
infrastructure to get in on the act, and the "ivo:" string means   `don't
ask me, find a resolver and ask it'.  (I'm not in the least   disputing
that the Registry has to be involved; the issue is how to   do it as
cheaply as possible).

That means that the question really is whether the costs of mandating   a
resolver are small enough to justify the benefits obtained.  Those  
costs, remember, include writing an RFC for "ivo:" and getting it  
approved by the IETF, designing an API and/or protocol for the
resolver(s), and writing the resolver servers themselves.  The
benefits of the indirection are that

   1. abstraction helps persistence by decoupling the identifier from
its retrieval
   2. the indirection means you `de-brand' the resolution -- you
can't easily detect who's serving you the actual resource.

Point (1) is less persuasive than it sounds: domain names like cds.u- 
strasbg.fr will change or fail on roughly the same timescale that  
resolution servers will change or fail, so that you haven't
fundamentally improved your reliability by the order of magnitude or   so
that would justify the extra costs.  (2) is only an advantage if   you
have no easy fallback.

This is where the ARK scheme that I mentioned seems to hit all the   sweet
spots, and seems to cheaply provide the identifier properties   we want.

ARK IDs are HTTP URLs and are, it seems, intended to be passed around   as
such.  However the "ark:.*" string within that URL is the
persistent bit, and can either be chopped out of the URL (if you want  
the de-branding for some reason), or chopped out and reattached to   the
resolver of your choice.

In some sense, therefore, ARKs are just like the ivo: identifiers,  
except that the resolution mechanism is explicitly specified as
`append this to the HTTP URL for a resolver service'.  There are a   few
other bells and whistles to do with the '?' and '??' stuff I   mentioned,
but that's basically it; extremely simple.  The cost of   this is that ARK
comparison becomes a bit more expensive (you have to   strip the resolver
prefix, or handle the suffix separately).

The benefits are that there's no RFC to write, no API to write, and   all
the costs are removed from the client end (so you can dereference   an ARK
using wget!) because you don't need any further tools (see  
<http://www.doi.org/tools.html>).  Plus you regain all the library,  
application and intelligibility benefits of using HTTP, and you get   BIND
and Apache to do the heavy lifting (getting someone else to do   the work
is always a good thing).

DOIs and Handles appear to be like ivo:, but are arguably like ARKs   in
many cases.  Pace bells and whistles, gongs, tubas and son et   lumiere in
the DOI API, you can basically resolve DOIs by appending   them to the URL
http://dx.doi.org (roll out the 80/20 arguments).

I quoted an RFC above.  The RFC persistent ID is a string of the form  
'^RFC[0-9]*$', but to resolve one, you can transform it into the HTTP  
URL http://www.ietf.org/rfc/<lowercased-RFC-number>.txt.  Conversely,  
given that URL, you can extract the document ID and use your own RFC  
resolver, or I can save you the trouble by quoting it convolved with   its
resolver.

In this last case, as with the ARKs, there is nothing _lost_ by
quoting it as an HTTP URL, but there's a lot that's gained in terms   of
simplicity and usability.  It means that

     % curl http://www.ivoa.net/ivo-resolver/ivo:/blah | xsltproc
grokVOTable.xsl | ...

...is a VO-enabled application.



OK, _I_ at least am more convinced than I was at the top of the
page.  How about you, Roy?

All the best,

Norman


-- 
------------------------------------------------------------------------ 
----
Norman Gray  /  http://nxg.me.uk
eurovotech.org  /  University of Leicester, UK






_______________________________________________
compute mailing list
compute@...
https://acooke.dyndns.org/mailman/listinfo/compute