[Policy] Mircea's suggestions

[David McTavish <dmctavish@sandvine.com>]

Title: RE: [Policy] RE: PCELS position

Some good suggestions here Mircea, but I believe that they will also inevitably require modifications to the underlying PCIMe document. Hopefully, we can pursue these options to find some common ground.

 

A.B. From first glance, I'm not sure about A or B, but if this could be implemented through "inferrence", then I'm all for it. However, reviewing PCIMe, it seems that the document intends to dictate the underlying implementation strategy via the class inheritence diagram. If the relationship between Rules/Groups and PolicySets can be abstracted out in PCELS, then I believe this would remove the necessity to deprecate pcimRule and pcimGroup objects. However, I am not sure if this was the intent of the document, and will leave for others to discuss, as obviously, I'm somewhat biased. :)

 

C. My opinion, from PCIMe is that C violates in its current form. In particular, I'd be concerned with the following quotations from PCIMe:

Section 3.2.3:

"Drawing from both QPIM and ICPM, the Priority property has been
   deprecated in PolicyRule, and placed instead on the aggregation
   PolicySetComponent.  ...  With the removal of the Priority property from
   PolicyRule, a new modeling dependency is introduced."

I believe if this was rephrased such that the last sentence permitted the existence of the Priority property within the PolicyRule object instead of calling for its outright removal, then the proposal would preserve room for backwards compatibility.

Section 5.5

"PolicySetComponent.Priority MUST have a unique value when compared with others defined for the same aggregating PolicySet"

This means that Priority MUST have a unique value for all rules/groups, which means existing PCIM implementations would be non-compliant. In order for this to be backwards compatible, it needs to be relaxed to allow for a default of zero (or non-existence), and process accordingly. I believe that PCIMe needs to be reworked in this capacity to have more leniency.

 

D. I'm not quite sure which direction you were inferring for the inheritence... are you inferring that a ReusableContainer object would extend from the current Repository object class?  This could be a more practical approach, but I still get the feeling that we are just patching an object to change the name, and the existing problem is maintained because the object dependency must be fully qualified within LDAP. (ie: a ReusableContainer object in ldap would have the following objectclasses defined {top, pcimPolicy, pcimRepository, pcimReusableContainer, pcimReusableContainerInstance}).

 

Regards,

d.

 

-----Original Message-----
From: Pana, Mircea [mailto:mpana@metasolv.com]
Sent: Monday, September 22, 2003 10:21 AM
To: 'Wijnen, Bert (Bert)'; 'David McTavish'; Pana, Mircea; 'policy@ietf.org'
Cc: 'John Strassner'; 'Joel M. Halpern'
Subject: RE: [Policy] RE: PCELS position

Maybe there is no need for such drastic measures. Maybe it is only a matter of interpretation of the PCIMe recommendations. After all PCIMe is quite lenient wrt. that is and what is not used in submodels (see PCIMe section 5.10.).

Some of the structural changes proposed by PCIMe make it difficult for PCELS to be interoperable with PCLS. These are as follows:

1. PCIMe defines a new abstract class, PolicySet, and makes it a superclass of the already defined PolicyRule and PolicyGroup

2. In PCIMe the PolicyRule.Priority property has been deprecated in favor of a new relative priority mechanism.
3. PolicyRepository is deprecated in favor of the new ReusablePolicyContainer.
 
PCELS could be interoperable with PCLS if it was to interpret these PCIMe changes as follows:
A. there is no need to have an explicit LDAP mapping of the abstract PolicySet. (see also B.)
B. there is no need to have an explicit LDAP mapping of the modified PolicyGroup. Implementations can use (the equivalent of) a PolicyRule with no Actions or Conditions for PolicyGroup objects.

C. implementations SHOULD (as opposed to MUST) use the relative priority mechanism instead of the absolute priority attribute of PolicyRule

D. PolicyRepository SHOULD not be used directly but it is acceptable for instances of this class to occur through inheritance.

So, the question is whether the statements A. through D. violate PCIMe or not. Opinions?

Thanks,
Mircea.
-----Original Message-----
From: Wijnen, Bert (Bert) [mailto:bwijnen@lucent.com]
Sent: Sunday, September 21, 2003 6:14 AM
To: 'David McTavish'; 'Pana, Mircea'; 'policy@ietf.org'
Cc: 'John Strassner'; 'Joel M. Halpern'
Subject: RE: [Policy] RE: PCELS position

W.r.t.
>  Is PCIMe considered so complete, that it is beyond modification, if such
>  modification could preserve its intent while also adhering to the desires
> of maintaining consistency with PCIM and PCLS?

PCIMe is at Proposed Standard. If, for example because of this effort to try and MAP it onto LDAP, we
find that we did some things in PCIMe that we should not have done, then, with WG consensus,
we can make incompatible changes to PCIMe and then recycle at Proposed Standard.
That is part of the normal standars track process. That is, we get something to PS, then we start
using/implementing (the "using" part is reusing PCIMe definitions in otehr CIM docs (like the
other docs we did in Policy, and like the IPsec work, the "implementing" is sort of mapping onto for
example LDAP I think)... and if we find major issues, then we fix and recycle at PS. If we do not
find major issues, we may advance to DS.

Hope this helps.
Bert