Hi Jensen,

Thanks for your answer. I was just resuming one use case that was motivating 
the need to disambiguate resource dependencies.

Would it be possible to look at the example where:
- entity = pid:MYPID
- property = FCI.capability10
Property definition depends on FCI map, entity ID depends on a Network map.

And write an example for:
- IRD entry for filtered/unfiltered propmaps,
- example request ,
- Example response

The purpose is to illustrate the problem and collect more WG feedback. Maybe 
the use case above does not exist but we may want to make sure we don’t miss 
other cases where both entity ID and property depend on an information resource.
Thanks,
Sabine



From: Jensen Zhang <[email protected]>
Sent: Thursday, April 11, 2019 6:35 PM
To: Randriamasy, Sabine (Nokia - FR/Paris-Saclay) 
<[email protected]>
Cc: IETF ALTO <[email protected]>; Richard Yang <[email protected]>
Subject: Re: Final Decision of Unified Properties Design before Moving to WGLC

Hi Sabine,

I remember that is the problem I proposed to motivate the design option 1. But 
in design option 2, we no longer have this problem. Let me clarify a little bit:

Why a property map requires dependencies? Because the client requires other 
resources to help it to understand the information on a property map. More 
specifically, the client wants to understand every key appearing in a property 
map. Those keys include entity identifiers and property names. Each entity 
identifier or property name may be defined in another resource (its origin). 
Without this resource, the client cannot understand the corresponding entity 
identifier or property name. That is one of the insights of design option 2.

In design option 2, we require the server to explicitly expose the origin of 
each entity identifier and property name to avoid ambiguity. But we notice that 
each entity identifier or property name has exactly one origin. I cannot come 
up with an example where an entity identifier or a property map has more than 
one origin.

In your PID-FCI example, if FCI capabilities are defined on PIDs, the map would 
depend on both Network Map and FCI map. But the Network Map is the origin of 
PIDs, and the FCI map is the origin of FCI capabilities. So each key still has 
one dependent resource.

I'm not sure if we will have an example where the entity identifier encoding is 
so complex that the client needs multiple information resources to parse this 
entity identifier correctly in the future. But so far, I cannot come up with 
such a real example. If we consider how to handle this, we may take a risk 
dragging on the overdesign.

Best,
Jensen


On Wed, Apr 10, 2019 at 6:01 AM Randriamasy, Sabine (Nokia - FR/Paris-Saclay) 
<[email protected]<mailto:[email protected]>>
 wrote:
Hi Jensen,

Thanks a lot for the provided examples. It will be indeed be helpful to present 
a fully fleshed example for the 2 options and the related pros & cons.
That is: example information resource in IRD, example request and response.

My question on option 2 and in general is to see how to handle examples where a 
property map depends on 2 or more resources.
For example, if FCI capabilities are defined on PIDs, the map would depend on 
both Network Map and FCI map.
Questions:
- does this example make sense?
- what is the probability of having similar cases of property maps depending on 
multiple other information resources?

Thanks,
Sabine



From: Jensen Zhang 
<[email protected]<mailto:[email protected]>>
Sent: Tuesday, April 09, 2019 4:28 PM
To: Randriamasy, Sabine (Nokia - FR/Paris-Saclay) 
<[email protected]<mailto:[email protected]>>
Cc: IETF ALTO <[email protected]<mailto:[email protected]>>; Richard Yang 
<[email protected]<mailto:[email protected]>>
Subject: Final Decision of Unified Properties Design before Moving to WGLC

Hi all,

Authors of the document draft-ietf-alto-unified-props-new had a discussion 
about the unified properties design last week. We reviewed two design options 
proposed in IETF 104 and analyzed the pros and cons of both.

For the design option 1, binding resource dependencies to property type, it is 
easy to process but hard to understand (we spend a lot of time trying to 
clarify the design).
For the design option 2, binding resource dependencies to each entity and 
property, it is easy to understand (analogous to the relational database) but 
hard to specify (e.g., IANA registry). Fortunately, authors already have a 
proposal about the IANA registry design of design option 2, which requires 
three new registries for entity domain types, properties, and resource types.

But we still need to make the final decision before we move forward.

Hi Sabine,

You mentioned that you still had some questions for the design option 2. Could 
you post them here? I started to revise the document based on the design option 
2, but have not merged it to the latest revision. I hope our co-authors can 
agree on a design at least before we moving to the document revising for WGLC.

There are some materials talking about two design options:

[1] 
https://datatracker.ietf.org/meeting/104/materials/slides-104-alto-unified-properties-for-alto-01.pdf
[2] 
https://docs.google.com/presentation/d/1lCcLLbyKqZjGADxcHSorfADKx_CoG1fz_j6GBfPGZQY/edit?usp=sharing

Best regards,
Jensen
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to