[ogsa-wg] [ogsa-naming-wg] ws-naming: Returning an alias-EPI in the referral-fault

[Frank Siebenlist]

Comments in-line for the discussion tomorrow:

Andrew Grimshaw wrote:
> Frank,
> We discussed this in our group this morning - and don't believe that the
> specification requires any changes to support the use cases. I'll explain
> below.
>
>   
>> -----Original Message-----
>> From: ogsa-wg-bounces at ogf.org [mailto:ogsa-wg-bounces at ogf.org] On Behalf
>> Of Frank Siebenlist
>> Sent: Thursday, September 14, 2006 10:24 AM
>> To: OGSA WG; ogsa-naming-wg at ggf.org
>> Subject: [ogsa-wg] ws-naming: Returning an alias-EPI in the referral-
>> fault
>>
>> I'd like to share two use cases for the return of a alias-EPI a an
>> alternative to returned referral resolver services:
>>
>> * Two EPIs are independently generated for the same resource.
>> If we associate a resource with some bio-object, the it happens that two
>> different researchers "name" the same object independently from each
>> other. When this is discovered, one needs the option to consolidate the
>> metadata and policy "underneath" one of those EPIs to lessen the burden
>> of maintenance and to ensure consistency. A common solution is to assign
>> a "master-EPI", and to let the other EPIs refer to that master-EPI when
>> it receives resolution requests.
>>
>>     
> [Andrew Grimshaw] 
> A couple of things. First, we thought about this and believe that it is
> really an upper level naming issue - there is a desire to have a human name
> that is re-mapped to a different entity - the two resources are DIFFERENT
> resources - the original - and the "new" one. Thus, by remapping an RNS name
> we can accomplish the goal.
>   

There are no human names here and RNS has no use.

> Assuming for the moment that the above does not satisfy your use case - then
> what you are trying to do is to redirect access to the "old" resource to the
> "new" resource. Your suggestion was to return a "fault" and a new EPI. We
> believe that is the wrong approach. Instead an EPR that points to the new
> "resource" should be minted that still contains the old EPI (if there was an
> EPI to begin with). Thus, to the client it is still the same resource --
> just at a different location. The client does not have to go through and
> change all of their references to the old EPI at all.  This way the
> semantics of what you are trying to accomplish are hidden away in your
> particular implementation - and not exposed at the semantic level.
>   

EPIs can be communicated in many ways and live forever - they can be
embedded in EPRs, but can also be communicated through queries, emails
or pigeons.

Assume that I have an EPI-1 that was issued for a certain resource and
the resolution info was maintained in resolution server R-1, and an
EPI-2 issued for another resource and that resolution info was
maintained in resolution server R-2.
Then it is discovered that those two resource "are" the same identical
resource
It is then decided that the resource's resolution&meta-data info is
consolidated and maintained in one location only, say R-2. Furthermore,
after the consolidation policy will be expressed on EPI-2 only.
When a global naming system is used for EPI-1, like dns/lsid/handle,
then that EPI-1 will always be bound to resolution server R-1.
In such cases, one would want R-1 to resolve to the new EPI-2 as a
referral to the EPI where the current resolution/metadata is maintained.

R-1 could potentially return a ReferenceResolver-EPR instead, which
would refer to the EPI-2 entry at R-2, and this would be allowed by the
current spec.

Semantically there is no difference between returning a referral through
an alias-EPI or a referral through a ReferenceResolver, and it is a
choice of the deployed naming server implementation which one makes sense.

Therefor there shouldn't be any objection to the suggested enhancement
especially when a real use case is identified.


>> * Moving metadata management to a different admin domain.
>> When a global naming system and global resolution frameworks are used,
>> like DNS, HandleSystem or LSID, then the EPI's URI will include
>> information about the naming server, i.e. the "prefix". This prefix in
>> the name is commonly mapped to a real, physical identifier service that
>> is used to administer the metadata/resolution bindings and such a server
>> will be part of a certain admin domain. When the owner/administrator/PI
>> of the individual identifier binding moves/changes-jobs, she often wants
>> to take the data-objects with her to the new admin domain as well as the
>> ability to administer the identifier bindings. Unless that user "owns" a
>> whole prefix, the administration of the identifier has to remain at the
>> original naming server. This is often not an acceptable solution both
>> for the naming server admin domain as well as the original identifier
>> owner. A common solution is to create a new EPI with a prefix that is
>> maintained in the new admin-domain, and to use that for all the
>> bindings/resolution information, and to add to the original EPI a
>> one-time referral to the new EPI.
>>
>>
>>     
> [Andrew Grimshaw] 
> I can see where you are coming from - especially having been deeply involved
> in LSID's. However, the WS-Name spec specifically says that the EPI should
> be location independent - and that clients should not assume anything about
> the internal structure (besides it being a URI). Embedding the name of the
> name server in the prefix is not something that implementations should count
> on (client implementations; if you're implementation wants to do that - fine
> - but it should not effect the overall specification.) Remember, that the
> client can uses any resolver to attempt to resolve the name - they don't
> have to use the one in the EPR.
>   

Not sure where we state that an EPI should be "location independent"...
or not even sure what that means in your context.

The spec should not rely on any specific resolver information embedded
in an EPI, but it should neither exclude any use of such information.

We have chosen the URI format for our EPIs for a reason (at least I
have...) as it allows the client applications to find the right
resolver. Your suggestion that a client can use "any resolver to attempt
to resolve the name" doesn't sound very efficient.  Our implementation
will make use of the fact that those EPIs are URIs, and the first
dispatching will be based on the URI-schema to find the right resolver
(uuid:/http:/dns:/lsid:/hdl:). Then depending on available
schema-specific handlers, one could associate resolver servers with
naming authorities. All those are "optimizations" that make sense if we
use real globally resolvable naming schema and I sure hope that we will
not cripple our spec to make it impossible to use those strategies.


>> The semantics for the referral fault is something like:
>>
>> "The resolver service that was asked for the resolution was unable to
>> provide the caller with the resolution information. However, the
>> resolution service has alternative information returned which the caller
>> may optionally use to try to resolve."
>>
>> It seems to me that returning a alias-EPI fits very well within the
>> referral fault semantics. Furthermore, the processing of a returned
>> alias-EPI seems completely equivalent to the processing of the original
>> EPI with some checks for looping and such.
>>
>> Enjoy the F2F without me.
>>
>> Regards, Frank.
>>
>> --
>> Frank Siebenlist               franks at mcs.anl.gov
>> The Globus Alliance - Argonne National Laboratory
>>
>> --
>>   ogsa-wg mailing list
>>   ogsa-wg at ogf.org
>>   http://www.ogf.org/mailman/listinfo/ogsa-wg
>>     
>
> --
>   ogsa-naming-wg mailing list
>   ogsa-naming-wg at ogf.org
>   http://www.ogf.org/mailman/listinfo/ogsa-naming-wg
>
>   

-- 
Frank Siebenlist               franks at mcs.anl.gov
The Globus Alliance - Argonne National Laboratory