"Bogaert, Bart (Nokia - BE)" <[email protected]> wrote:
> "Bogaert, Bart (Nokia - BE)" <[email protected]> wrote:
> > On Tue, Sep 06, 2016 at 07:50:47AM +0000, Bogaert, Bart (Nokia - BE)
> wrote:
> > > 
> > > [Bart Bogaert] When sending a configuration request to a device 
> > > while there is no HW physically present yet is what we call 
> > > pre-provisioning meaning that the configuration is made up-front in 
> > > attendance of the HW being plugged at a later stage.  When the 
> > > plugged HW does not meet the pre-configured data I think it is 
> > > normal that an alarm is raised but that does not take away the fact 
> > > that the device was configured in
> > advance.
> > > 
> > 
> > In your example, there is nothing configured as far as I can tell.
> > 
> > The way the interfaces data model supports pre-configuration is by 
> > having a _name binding_; a pre-configured interface is applied once 
> > the name of the pre-configured interfaces matches the name of a
> > (physical) interface. I think Martin is asking the question whether 
> > the same model of using name bindings can be applied in your case and if
> not why not.
> > 
> > [Bart Bogaert] I'm afraid I am lost in the names being used here.  
> > Whether it's called name binding or pre-configuration or still 
> > something else, the key point is that we configure objects in the 
> > device prior to them being physically present, nothing more, nothing 
> > less.  The consequence of this being that these objects should be 
> > present in the data tree of the device and once the HW gets plugged 
> > all operational data linked to that comes into existence too.  I do 
> > not know of another way to explain what is intended with what we call
> 'pre-configuration'.
> 
> I think there are two things that I am trying to understand:
> 
>   (i) Identification
> 
>       In order to be able to pre configure a component, the operator
>       needs a deterministic way to idenfity the component.  How
>       exactly is this done?
> 
>       One way is to rely on the name of the to-be-present component.
>       Another way is to identify its parent by name, and provide a
>       parent-rel-pos integer.  (The latter assumes that the parent's
>       name is known...)
> 
> [Bart Bogaert] If we stick to the use case of equipment the
> pre-configuration is based on containment and names attributed to the
> entities by the operator.  I tried to explain that so far names attributed
> to entities have no other meaning then identifying the entity and, as
> allocated by the operator, there is nothing to be predicted: it is
> known.

Dut even in your example of simple containment, it works b/c the
operators knows the name of the parent, and the parent-rel-pos.
Right?  So what if you need to pre-configue two levels of containment?
You know the name of the top-level component, so you can pre-configure
the first child.  But then you'll have to know the name of this first
child in order to pre-configure the grand child.

Or are you saying that the system always creates and gives names to
all top-level components, and then the operator or system can given
name to sub-components?

> I
> agree that in the case of entities there will be no other special leafs that
> require to be configured.
> 
>   (ii) Configuration
> 
>       Once a component is identified, what parameters can you
>       pre-configure?
> 
>       One answer could be "none", which I think is what the examples
>       you have provided so far have.  This then would not really be
>       pre-configuration, but rather a constraint, as Juergen
>       indictated.
> 
> [Bart Bogaert] As indicated above that might be true for equipment (entity)
> objects but pre-configuration could also include definition on top of these
> pre-configured ports and in this case there will certainly be configuration
> of leafs that are characteristic for the to-be offered service and which
> will be defined on top of a port (hence the augment of interfaces with
> back-pointer to the port entity).  Also in this case the operator knows
> exactly on top of which port the interface will be created.  The name
> attributed to the interface has no special meaning to the SW of the device
> apart from giving it a unique reference.  The name is set by the network
> operator and can basically be anything (that is why I say that it has no
> special meaning to the SW in the device).  As the names are set by the
> operator there is nothing that requires prediction.  The name-binding as it
> is referred to here comes from the model where the leafrefs point to the
> name of underlying resource (be it entity or interface).

I have a hard time trying to understand what you actually need.  It
now seems that you want to pre-configure some physical entity in order
to be able to refer to it from higher-level models.  Is this correct?
Thus, you don't really want to pre-configure any leafs in the entity
model, except for the name.

Remember that the thread started with a request to have the 'class' and
'contained-in' leafs in the config list physical-entity, and 'class'
being mandatory.

> Maybe we have a different understand of "pre-configuration"?  For me it
> simply means that an operator is able to create a configuration in the
> device without the HW supporting this configuration being physically present
> but is getting stored in the database of the device (I'm explaining it as it
> is in our SNMP-based devices).  The configuration gets applied once the
> supporting HW is plugged in the device (given the fact that the plugged HW
> matches what was previously configured).

Right, this is also my understanding of pre-configuration.


/martin

_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod

Reply via email to