Hi Michael and Qin,
Thanks for the constructive discussions.
In line...
> > I have quickly scanned this document. Some initial comments [ms]:
> >
> > *** Page 3:
> >
> > Service Model: A service model is a specific type of data model.
> >
> > It describes a service and all of the parameters of the service in a
> > portable, operator-independent way. It can be used by a human or
> > by software to configure or request a service and may equally be
> > consumed by a human or by a software component.
> >
> > [ms] I would suggest a more careful wording regarding the completeness
> > of the model, e.g., "It aims at describing a service and all of the
> > parameters of the service...". As discussed in L3SM in the past, for
> > certain services (e.g., firewalls, DoS prevention, ...) the service
> > model would have to be extended / augmented. Also note that the
> > statement "all parameters" contradicts the statement on SLAs later in
> > the text.
> >
> > [Qin] What the service is has been well defined and clarified before
> > service model definition in the section 2.
> > That is to say, Firewall and DoS prevention are just additional
> > functionalities that are used to enhance the basic service we defined
> > in the document.
> > For what the service is not is further clarified in the section 5, the
> > first bullet.
>
> [ms2] I am not sure if I can follow. My point is that the wording "describes a
> service and all of the parameters" seems contradicting given the use of the
> term
> "all" if you exclude SLAs. Also, Section 5 does not really have a well-defined
> definition - excluding some examples does not necessarily result in a preciser
> definition. (I agree that "Service" is challenging to define but maybe this
> could be
> one reason to find an alternative term to "service model").
>
> [Qin]: As we clarified Service Level Agreements (SLAs) have a high degree of
> overlap
> with the definition of services present in service models, but we don't want
> to
> cover a number of fine-grained details of attributes defined in SLA for the
> reason
> given in section 5.
> That's why we add a constraint after " describes a service and all of the
> parameters " to limit these parameters to abstract level and operator
> independent way. Maybe we could add
> some text to make this more clear.
I think this is a language issue.
I tend to agree with Michel that "all" is a risky word.
Clearly we (well, OK, I wrote those words) meant "all of the parameters that
describe the part of the service we are describing". We did not mean *all* and
that should be clear from the discussions of augmentations that operators might
make to cover their own additions to the base service.
What we are trying to do is:
- define a core, base service that all operators agree on
- list *all* of the parameters of that core, base service
- list all of the settings of those parameters necessary to define
that core, base service.
We can make that clearer.
> > In addition, we don't take SLA fine granularity details combining
> > commercial terms as service parameters, so I don't see there is
> > inconsistency issue, for some misconception on the service model
> > Understanding is clarified in the different section, this is
> > intentional.
>
> [ms2] My point is that SLA parameters may be an important part of the
> customer-
> service provider agreement of a service and it is surprising to first seeing
> this
> mentioned at the end of Section 5.
>
> [Qin]: Sure, to avoid confusion with the service defined in the model, as I
> mentioned above, we use the different term for these Quality related
> parameters.
I agree with both of you here.
Without doubt, SLA will form an important part of the commercial agreement
between operator and customer.
However, with the exception of some high-level parameters of the service (such
as bandwidth) we find that SLA parameters vary widely between operators.
I'm aware that a number of organisations (most notably TMF) have spent a lot of
time parameterising and generalising SLAs. This may be something we can try to
do as well, but in my opinion it is both hard and an extension to the service
model work we are doing.
> > [ms] I find the wording "it can be used by a human..." confusing. Is
> > the assumption that a user would manually type in a service request
> > resulting in a valid data model, e.g., by manually typing in NETCONF
> > commands? I would rather write something like "the parameters can be
> > entered by software or also manually by a human..."
> >
> > [Qin]: what we try to convey here is Human can rely on CLI,script or
> > other human readable language to define a service, request the
> > service, but such parameter will not directly implemented by network
> > element.
>
> [ms2] Sure, but what is the role of a service model defined in YANG e.g. when
> the human relies on CLI or scripts? To me this wording does not explain this
> well.
>
> [Qin]: why we should tie service model definition with YANG, the service
> model is
> just a service template, you can use human readable language to express it or
> you
> Can use machine readable language to express it, it doesn't matter.
> Therefore, in
> some cases, you just need fill the service parameters in some kind of form and
> submit this form to the server for further parsing and interpretation.
I think we might go for "it can be used by a human via a user interface such as
a GUI, web form, or CLI" to cover the meaning.
What we are really doing is distinguishing between human control (by whatever
means) and application (i.e., autonomous software) control.
> > *** Page 3:
> >
> > The encoding and communication protocol used to exchange a service
> > model between customer and network operator are deployment- and
> > implementation-specific. The IETF recommends the use of the NETCONF
> > Configuration Protocol [RFC4741] with data encoded in XML or JSON
> > for
> > interactions "on the wire" between software components. However,
> > co-
> > located software components might use an API, while systems with
> > more
> > direct huan interactions might use web pages or even paper forms.
> >
> > [ms] I am not sure if the IETF should suggest the use of NETCONF on
> > the northbound interface of a "service orchestrator", which seems to
> > be the scope of this document. Are there pointers to running code that
> > actually do that? I would rather prefer a wording such as "The IETF
> > has standardized NETCONF and RESTCONF for interactions "on the wire"
> > between software components".
> >
> > [Qin]: You are right, I think RESCONF is recommended on the northbound
> > interface of a service orchestrator, your proposed change looks good.
> > Thanks.
> >
> > *** Page 7:
> >
> > o The network operator may use further data models that help to
> > describe how the service is realized in the network. These
> > models
> > might be used on the interface between the Service Orchestrator
> > and the Network Orchestrator as shown in Figure 3 and might
> > include many of the pieces of information in the service model
> > alongside protocol parameters and device configuratin
> > information.
> > It is important that the Service Orchestrator should be able to
> > map from a service model to these data models, but they are not
> > the same things.
> >
> > [ms] Why is it not possible that the data models are the same? In
> > Figure 3, the "service orchestrator" is actually a "client" (read this
> > as "customer") of the "network orchestrators". Inside an operator, the
> > "service orchestrator" could be operated by a different organization
> > than the "network orchestrator", i.e., there ould even be a business
> > relationship. In general, architectures can be recursive. I think this
> > statement requires further explanation or rewording.
> >
> > [Qin]: The concept to split Orchestrator into service orchestrator and
> > network orchestrator, is the service orchestrator is network agnostic
> > while network orchestrator is network aware.
> > Service model consumed by the service orchestrator could be different
> > from the data model consumed by network orchestrator since network
> > orchestrator need to decide what network device is selected, which
> > protocol is enabled, which technology is used, which network interface
> > or port is assigned. The service orchestrator doesn't need to care
> > about the underlying technology or what protocol is available.
>
> [ms2] You may want to add that assumption more explicitly to the text and
> possibly add that this is only one potential architecture.
>
> [Qin]: We give two optional architecture in the figure 2 and figure 3, let us
> know if
> there is any potential architecture we are missing.
Additionally, we tried to capture this in Section 5.
> > *** Page 7:
> >
> > o Service Level Agreements (SLAs) have a high degree of overlap
> > with
> > the definition of sevices present in service models. Requests
> > for
> > specific bandwidth, for example, might be present in a service
> > model, and agreement to deliver a service is a commitment to the
> > description of the service in the service model, however, SLAs
> > typically include a number of fine-grained details about how
> > services are allowed to vary, by how much, and how often. SLAs
> > are also linked to commercial terms with penalties and so forth,
> > and so are also not good topics for standardization.
> >
> > [ms] I wonder if it could make sense to distinguish between SLA and
> > Service Level Objective (SLO) or Service Level Requirement (SLR).
> > While I agree that it is difficult to standardize commercial terms,
> > this IMHO partly contradicts the positioning of the term "service
> > model" earlier in the document. Maybe the scope of the definition of
> > "service
> model"
> > has to be narrowed down?
> >
> > [Qin]: We have clarified what the service is in the section 2 and Our
> > service model definition aligns with service definition. Yes, we need
> > to distinguish SLA from Service Level Objective/Requirement, but not
> > necessary introduce another new term. We have clarified their
> > difference in the text you quoted.
>
> [ms2] Well, I think you should comment on SLAs in Section 2 to avoid
> confusion.
>
> [Qin]: Done, see above.
>
> > *** page 9:
> >
> > 6.4. Supporting Multiple Services
> >
> > Network operators will, in general, offer many different services to
> > their customers. Each would normally be the subject of a separate
> > service model.
> >
> > [ms] The actual question here is if there could be communality between
> > service models for similar services. That problem cannot be solved by
> > this document but the text could e.g. explain that for a customer it
> > can be confusing if separate service models model related concepts in
> > different ways.
> >
> > [Qin]: That is a good question, I hope some building blocks specified
> > by one service model can be reused in another service model.
> > Unfortunately we don't have too many YANG models that are used to
> > model service in the IETF. We could cover this in the next version if
> > more such service models are needed.
Agree with both of you.
We're still playing with our first few service models so it is hard to guess
how this will pan out. But I suspect that the "building block" approach will be
what we end up with.
Best,
Adrian
_______________________________________________
OPSAWG mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/opsawg
