registry of registries

[Aurelien Stebe]

The idea sounds good to me too, but I read it in a different way.

The master IVOA registry would contain "registry" type resources 
exclusively.
A new registry (publishing or full) would send his personal "registry" 
resource metadata
to this master registry. Then, harvesting registries would always start 
their harvesting
cycle by a query to the master registry to see if there is any new 
registry around to harvest.
This way, harvesters, if doing a complete cycle, could request for only 
managed resources,
since they would be sure to know about all existing registries; hence, 
no need to sort harvest results.

With this aproach, we should discuss if the master registry need to know 
about each registry's managedAuthorities
(because then for each new managedAuthority, the registry would need to 
update the master registry infos;
but authorities conflicts could be easily detected and even be made the 
master registry's responsability)
and the addition of info to registry records like : is it publishing ? 
full ? supporting sets ? specialised in some domain ?

Your way of seeing things Kevin makes life easier for full registries, 
but I think this would put a
big load on the master registry and make it a critical single point of 
failure.
Whereas with the senario I mentioned, if the master registry came to 
fail, the harvest of the
new registries would only be pushed back by the time needed to fix it.
All the rest would keep working just fine.

cheers,
Aurelien

KevinBenson wrote:

>This does sound pretty good, could you elaborate a little more on some
>scenarios just to make it clear exactly what is happening.
>
>----
>Let me go off on a little bit of a tangent to see if I migth have read it
>correctly:
>Could we use the master registry as the main harvester and harvesee?
>Meaning much like today(with full registries) the master registry goes off
>harvesting all Publishing and Full Registries quite frequently.  Then Full
>Registries could come and harvest the master Registry for everything it ever
>needs.  This sounds really good where Full Registries do not need to go to
>all the registries to get the same data that it could get at the Master
>Registry.  Where by something like oai:ListRecords always returns everything
>but normally via "from" date would be used.  Also assuming no search
>interface to the master registry.
>
>Cheers,
>Kevin
>
>-----Original Message-----
>From: owner-registry at eso.org [mailto:owner-registry at eso.org]On Behalf Of
>Robert Hanisch
>Sent: 07 April 2005 21:51
>To: registry at ivoa.net
>Subject: registry of registries
>
>
>Following today's regular NVO technical telecon, Ray Plante asked that I
>post a suggestion I made this morning to this list that might help
>clarify/simplify some of the fairly complex ideas people are discussing
>regarding hierarchies of registries and such.
>
>I would not expect that we will have a vast number of publishing registries.
>Dozens, perhaps hundreds, but not likely anything approaching the many
>thousands of resources we already have in the registries themselves.  Thus,
>we could easily track the publishing registries through an IVOA master
>registery.  A site with a publishing registry would register it in the IVOA
>master registry.  Sites providing full searchable registries could query the
>master registry to make sure their harvesting is complete.  There would be
>no need to learn of other registries via word of mouth, or through more
>complex or ad hoc approaches, and no need to worry strictly about national
>or international hierarchies.  This seems to me a natural function for the
>IVOA.  A replica site could certainly be set up to avoid single point of
>failure issues.  However, even this does not seem so important given that
>the only likely users of the IVOA master registry would be other registries,
>and there is little here that is of high time criticality.
>
>Folks on the NVO telecon seemed supportive of this idea.
>
>Cheers,
>Bob
>
>
>
>  
>