Re: [I2nsf] Should we call "South Bound Interface" for the interface between "controller <-> NSF", and "North Bound Interface" for the interface between "Client <-> controller"?

[John Strassner]

Hi,

I think it is better to have crisp and clear terminology, and make the
changes now, rather than wait and end up with ambiguous or confusing
documents.

While your clarifications are helpful, I don't think that they address my
concerns, and they definitely don't address Diego's concerns (which I
wholeheartedly agree with). Furthermore, I don't see what absolute
coordinates (north vs. south) really buys you. What if you have stacked
controllers (e.g., a controller that is managing multiple controllers)?
Finally, what if you don't have a controller, but some other entity?

best regards,
John

On Sun, Jun 26, 2016 at 3:41 AM, Loa Andersson  wrote:

> Folks,
>
> I think this comment would have been fine, and I guess I supported it if
> it would have been here two years ago. As it is now I think there are to
> much water under the bridges, and actually to many documents involved.
> We have a common usage of "North Bound" or "Northbound" and South Bound"
> or Southbound". The pain in changing is much greater than the gain.
>
> However, I would support adding text to the terminology section like
> this:
>
> North Bound Interface - the interface between the client and the
> controller
>
> South Bound Interface - the interface between the controller and the NSF
>
> /Loa
>
> On 2016-06-26 04:02, John Strassner wrote:
>
>> Hi Linda,
>>
>> During the I2NSF early stage (before the WG was created), "capability
>> interface" was used to represent the interface between controller <->
>> NSF, and "service interface" was used to represent the interface between
>> the Client <-> controller. 
>>
>> 
>> The term "capability interface" doesn't bother me. However, I don't
>> think that we are using it to its full potential - see below.
>>
>> The term "client interface" does bother me, because
>> 1) to me, it implies a client-server architecture (or a 3-tier
>> architecture on a stretch), and I don't understand why we should be
>> limited to that
>> 2) what is the client? To me, a client is a host. Aren't we talking
>> about the application here? The examples in your figure are arguably
>> SERVERS,
>> not CLIENTS. :-)
>> >
>> As many people use the terminologies loosely, the "Capability Interface"
>> being interchangeably used with "Capability Layer", and "Service
>> Interface" being interchangeably used with "Service Layer". 
>>
>> 
>> Warning, this is probably just me, but...
>>
>> ...I do NOT like "layers", because abstractions should work for
>> everything, not just sometimes. Look at our "layers" - we repeatedly
>> violate the true notion of a layer (e.g., MPLS). Look at all of the
>> "inter-layer" compatibility problems we've had over the years. Why do we
>> need to use layers in I2NSF? I would strongly argue to use "interface";
>> if that is not acceptable, then choose a different term (e.g., planes)
>> that does not have the traditional limitations of layers.
>> 
>>
>> The I2NSF Terminology Draft has defined the "Capability Layer"
>> (independent of which interface to the controller) for exposing the
>> capability of a domain (over Client Facing   Interface), or for exposing
>> the capability of a NSF (over the NSF Facing Interface). 
>>
>> By this definition, ECA Policy’s "Event" capability can be discovered
>> independently from the "Condition" capability, or "Action" capability.
>>
>> 
>> Sorry, I disagree.
>>
>> Events, conditions, and actions are NOT capabilities. Capabilities are
>> defined in the Capabilities Draft as:
>>
>> Capabilities are functions that NSFs can perform. This interface
>> is used to advertise, select, and activate capabilities of selected
>> NSFs in a vendor-independent manner.
>>
>> Put another way, Capabilities are used to define functions that MAY be
>> used. There is no guarantee that they will be actually used.
>> Furthermore, Events, Conditions, and Actions are used to construct an
>> ECA policy rule. Events, Conditions, and Actions are types of Boolean
>> statements (at least in ECA rules) and have nothing to do with
>> Capabilities.
>>
>> While you MAY be able to relate a rule to Capabilities, they are two
>> completely separate things.
>> 
>>
>> Therefore, continue using the  “Capability Interface” can cause more
>> confusion in the future as its sound is too close to the “Capability
>> layer”.
>>
>>
>> 
>> Agree. So let's get rid of Capability **layer**. It isn't a layer,
>> because...
>>
>>
>> ...wait for it...
>>
>>
>> ...Capabilities could be used to describe NSF functions as well as
>> Controller functions. Thus, there is no "layer" in the classical
>> definition of the term "layer".
>> 
>>
>> __ __
>>
>> Therefore, we are asking people to state which of  the following options
>> should be used:
>>
>> __ __
>>
>> __1.  __Use “Client Facing Interface” for "Client <-> controller";
>> and “NSF Facing Interface” for "controller <-> NSF", 
>>
>> __2.  __Use “Controller North Bound Interface” 

Re: [I2nsf] Should we call "South Bound Interface" for the interface between "controller <-> NSF", and "North Bound Interface" for the interface between "Client <-> controller"?

[John Strassner]

Hi Linda,

During the I2NSF early stage (before the WG was created), "capability
interface" was used to represent the interface between controller <-> NSF,
and "service interface" was used to represent the interface between the
Client <-> controller.


The term "capability interface" doesn't bother me. However, I don't think
that we are using it to its full potential - see below.

The term "client interface" does bother me, because
1) to me, it implies a client-server architecture (or a 3-tier
architecture on a stretch), and I don't understand why we should be limited
to that
2) what is the client? To me, a client is a host. Aren't we talking
about the application here? The examples in your figure are arguably
SERVERS,
not CLIENTS. :-)

Warning, this is probably just me, but...

...I do NOT like "layers", because abstractions should work for everything,
not just sometimes. Look at our "layers" - we repeatedly violate the true
notion of a layer (e.g., MPLS). Look at all of the "inter-layer"
compatibility problems we've had over the years. Why do we need to use
layers in I2NSF? I would strongly argue to use "interface"; if that is not
acceptable, then choose a different term (e.g., planes) that does not have
the traditional limitations of layers.


The I2NSF Terminology Draft has defined the "Capability Layer" (independent
of which interface to the controller) for exposing the capability of a
domain (over Client Facing   Interface), or for exposing the capability of
a NSF (over the NSF Facing Interface).

By this definition, ECA Policy’s "Event" capability can be discovered
independently from the "Condition" capability, or "Action" capability.


Sorry, I disagree.

Events, conditions, and actions are NOT capabilities. Capabilities are
defined in the Capabilities Draft as:

Capabilities are functions that NSFs can perform. This interface
is used to advertise, select, and activate capabilities of selected
NSFs in a vendor-independent manner.

Put another way, Capabilities are used to define functions that MAY be
used. There is no guarantee that they will be actually used. Furthermore,
Events, Conditions, and Actions are used to construct an ECA policy rule.
Events, Conditions, and Actions are types of Boolean statements (at least
in ECA rules) and have nothing to do with Capabilities.

While you MAY be able to relate a rule to Capabilities, they are two
completely separate things.


Therefore, continue using the  “Capability Interface” can cause more
confusion in the future as its sound is too close to the “Capability
layer”.



Agree. So let's get rid of Capability **layer**. It isn't a layer,
because...


...wait for it...


...Capabilities could be used to describe NSF functions as well as
Controller functions. Thus, there is no "layer" in the classical definition
of the term "layer".




Therefore, we are asking people to state which of  the following options
should be used:



1.  Use “Client Facing Interface” for "Client <-> controller"; and “NSF
Facing Interface” for "controller <-> NSF",

2.  Use “Controller North Bound Interface” for "Client <-> controller";
and “Controller South Bound Interface” for “controller <-> NSF", or



Or you can provide a better option.




I choose option 3. :-)


The problem with "Client-Facing Interface" is that I'm not sure what a
"Client" is in NSF.

NSF-Facing Interface is OK; my problem is, why are we introducing Yet
Another Term?


The problem with Northbound and Southbound is that there is no clear
"north" and "south" here. Look at all of the projects that propose various
data models at both the device interface level AND the network management
application layer. So tell me, which is "south" here? :-)


Now, as for option 3, I'm thinking about it. However, I do think that you
have spotted an important inconsistency, so let's take time to fix it and
not rush into rash decisions.



best regards,

John



On Thu, Jun 23, 2016 at 3:31 PM, Linda Dunbar 
wrote:

> I2NSF WG:
>
>
>
> Need your opinion for a good name to represent “Client Facing Interface”
> and “NSF Facing Interface” of the I2NSF reference model:
>
>   +-+
>
>   |  I2NSF Client   |
>
>   | E.g. Overlay Network Mgnt, Enterprise network Mgnt  |
>
>   |  another network domain’s mgnt, etc.|
>
>   +--+--+
>
>  |
>
>  |  Client Facing Interface
>
>  |
>
>+-+---+
>
>|Network Operator mgmt|   +-+
>
>| Security Controller | < - > | Developer’s |
>
>+--+--+  Registration | Mgnt System |
>
>   |