[netmod] ECA Policy: What is an adequate abstraction level to express policies and intent

[Qin Wu <[e-mail]>]

Hi, Juergen:
Come back to the key issues for ECA Policy.
I think Policies need to be readable and hence be expressed at a high level of abstraction and in a suitable _language_.
But I propose to separate policy expression and intent representation, especially high level service intent representation, translation, mapping which is the hot topic in NMRG.
High level language we select for policy representation is YANG, expressed by the NMS or controller.
We can compare YANG vs NETCONF with RMON vs SNMP, RAMON is an extension of SNMP, provides traffic flow data for troubleshooting and the controls necessary to adjust for better performance from a central console.
I think we set similar goal as ROMON in our draft.

Unlike other intent translation or mapping, we compile High-level policy expression down into more verbose primitive representations that are closer to an execution abstraction. YANG expression is capable for such a compilation;.
Primitive representation in the device is script language, typical examples are Python or TCL used in the device

Of course there is pitfall to start somewhere in the middle of several layers of abstraction and then getting stuck somewhere when compiling things down to _efficient_ instrumentations.
One of lesson we learn from the past is SUPA is defined and described very abstractly, with its high-level blocks and UMLs, and is very difficult to be written in YANG and harder to be implemented.
We will avoid such pitfall.

At the current stage YANG is used for abstraction and representation. YANG is both representative and implementable.

-Qin (on behalf of authors)
-----邮件原件-----
发件人: netmod [mailto:[e-mail]] 代表 Juergen Schoenwaelder
发送时间: 2020年12月30日 2:56
收件人: Adrian Farrel <[e-mail]>
抄送: 'NETMOD Group' <[e-mail]>
主题: Re: [netmod] Adoption poll for draft-wwx-netmod-event-yang-10

Adrian,

some key issues when it comes to policy-based management systems:

 - What is an adequate abstraction level to express policies and intent?

   This question has no simple answer. I believe policies need to be
   readable and hence they need to be expressed at a high level of
   abstraction and in a suitable _language_. High-level policy
   expression may be compiled down into more verbose primitive
   representations that are closer to an execution abstraction. A
   common pitfall is to start somewhere in the middle of several
   layers of abstraction and then getting stuck with something awkward
   to put a clean higher layer abstract onto and to compile things
   down to _efficient_ instrumentations.

 - Where are policies executed?

   This can range from a logically centralized policy execution
   engine, which is part of what people call an orchestrator these
   days, to fully distributed policy execution models. In reality, you
   likely want to distribute functions dynamically but this makes
   solutions technically much more complicated. Given today's scalable
   computing and networking capabilities, logically centralized
   solutions are on the rise and have replaced the distributed
   approaches of the 90s.

 - When to detect and resolve policy conflicts?

   Detecting and resolving conflicts in larger collections of policies
   is non-trivial. This includes problems ranging from micro timescale
   atomicity issues to larger timescale stability issues (interacting
   policy control loops). If policy execution is distributed (or the
   event / information sources are distributed), this ultimately
   resolves to problems such as taking consistent snapshots or finding
   ways to work with inconsistent observations of a distributed system
   that are guaranteed to converge to stable states (self-stabilizing
   algorithms).

 - Who is interested in interoperable policy representations / languages?

   The IETF is about interoperability. What are the business models
   that push for interoperable policy based management standards? Who
   benefits from having an interoperable standard and how much effort
   are organizations willing to invest into engineering a reasonable
   solution addressing the other (non-trivial) questions raised above?
   Will they be implementing the solution in their products?

My position is that there are way too many difficult technical issues to resolve for this work to be viable for the IETF. Instead, I suggest that people go and work out solutions and once the silver bullet has been found, bring it to the IETF. (Historically, all attempts to cast policies into existing data models such as MIB modules or LDAP schema led to something awkward and unusable. I believe YANG modules are no
different.)

/js

Some relevant RFCs (there may be more):

3052 Service Management Architectures Issues and Review. M. Eder, S. Nag.
     January 2001. (Format: TXT, HTML) (Status: INFORMATIONAL) (DOI:
     10.17487/RFC3052)

3084 COPS Usage for Policy Provisioning (COPS-PR). K. Chan, J. Seligson,
     D. Durham, S. Gai, K. McCloghrie, S. Herzog, F. Reichmeyer, R.
     Yavatkar, A. Smith. March 2001. (Format: TXT, HTML) (Status:
     HISTORIC) (DOI: 10.17487/RFC3084)

3159 Structure of Policy Provisioning Information (SPPI). K. McCloghrie,
     M. Fine, J. Seligson, K. Chan, S. Hahn, R. Sahita, A. Smith, F.
     Reichmeyer. August 2001. (Format: TXT, HTML) (Status: HISTORIC)
     (DOI: 10.17487/RFC3159)

3318 Framework Policy Information Base. R. Sahita, Ed., S. Hahn, K. Chan,
     K. McCloghrie. March 2003. (Format: TXT, HTML) (Status: HISTORIC)
     (DOI: 10.17487/RFC3318)

3460 Policy Core Information Model (PCIM) Extensions. B. Moore, Ed..
     January 2003. (Format: TXT, HTML) (Updates RFC3060) (Status:
     PROPOSED STANDARD) (DOI: 10.17487/RFC3460)

3644 Policy Quality of Service (QoS) Information Model. Y. Snir, Y.
     Ramberg, J. Strassner, R. Cohen, B. Moore. November 2003. (Format:
     TXT, HTML) (Status: PROPOSED STANDARD) (DOI: 10.17487/RFC3644)

3198 Terminology for Policy-Based Management. A. Westerinen, J.
     Schnizlein, J. Strassner, M. Scherling, B. Quinn, S. Herzog, A.
     Huynh, M. Carlson, J. Perry, S. Waldbusser. November 2001. (Format:
     TXT, HTML) (Status: INFORMATIONAL) (DOI: 10.17487/RFC3198)

4011 Policy Based Management MIB. S. Waldbusser, J. Saperia, T. Hongal.
     March 2005. (Format: TXT, HTML) (Status: PROPOSED STANDARD) (DOI:
     10.17487/RFC4011)

4104 Policy Core Extension Lightweight Directory Access Protocol Schema
     (PCELS). M. Pana, Ed., A. Reyes, A. Barba, D. Moron, M. Brunner.
     June 2005. (Format: TXT, HTML) (Updates RFC3703) (Status: PROPOSED
     STANDARD) (DOI: 10.17487/RFC4104)

8328 Policy-Based Management Framework for the Simplified Use of Policy
     Abstractions (SUPA). W. Liu, C. Xie, J. Strassner, G. Karagiannis,
     M. Klyus, J. Bi, Y. Cheng, D. Zhang. March 2018. (Format: TXT, HTML)
     (Status: INFORMATIONAL) (DOI: 10.17487/RFC8328)

WGs/RGs that at least partially related to policy-based management:

- Simplified Use of Policy Abstractions WG (supa) (2015 - 2017)

- Policy Framework WG (policy) (1998 - 2004)

- Resource Allocation Protocol WG (rap) (1997 - 2005)

- Distributed Management WG (disman) (1996 - 2006)

- Services Management RG (smrg) (2019? - 2001?)

- Network Management RG (nmrg)

  - draft-clemm-nmrg-dist-intent (2017-2019)
  - draft-irtf-nmrg-ibn-concepts-definitions-02.txt (2019-2020)

Other resources:

- https://en.wikipedia.org/wiki/Policy-based_management

- https://www.youtube.com/watch?v=E_v-of582xg

- Boutaba, R. and I. Aib, "Policy-Based Management: A
  Historical Perspective". Journal of Network and Systems
  Management (JNSM), Springer, Vol. 15 (4), December 2007.
  https://doi.org/10.1007/s10922-007-9083-8

- Pavlou, G., "On the Evolution of Management Approaches, Frameworks
  and Protocols: A Historical Perspective". Journal of Network and
  Systems Management (JNSM), Springer, Vol. 15 (4), December 2007.
  https://doi.org/10.1007/s10922-007-9082-9

- Strassner, J., "Policy-Based Network Management: Solutions for the
  Next Generation", Morgan Kaufmann, December 2003.


On Tue, Dec 29, 2020 at 04:26:12PM -0000, Adrian Farrel wrote:
> Hi Juergen,
> 
> What you say about learning lessons from the past is wise and valuable.
> 
> Sadly (well, it's a good thing, really) we have new people in the IETF 
> and the memory of events over the last 20 years are not immediately 
> accessible to them. Others, who are old and grey, have been around 
> that long but were not necessarily involved in previous ECA discussions.
> 
> Since "intent-based networking" is a big thing once again (see recent 
> reports of acquisitions in this sector) the excitement about ECA may 
> be forgiven, but it would help to ground the discussions if those who 
> can remember previous efforts would share their experiences or at 
> least some pointers.
> 
> Best,
> Adrian
> 
> -----Original Message-----
> From: netmod <[e-mail]> On Behalf Of Juergen 
> Schoenwaelder
> Sent: 23 December 2020 18:09
> To: Andy Bierman <[e-mail]>
> Cc: NetMod WG Chairs <[e-mail]>; NETMOD Group 
> <[e-mail]>
> Subject: Re: [netmod] Adoption poll for draft-wwx-netmod-event-yang-10
> 
> On Wed, Dec 23, 2020 at 07:05:44AM -0800, Andy Bierman wrote:
> > On Wed, Dec 23, 2020 at 3:14 AM tom petch <[e-mail]> wrote:
> > 
> > > From: netmod <[e-mail]> on behalf of Dhruv Dhody < 
> > > [e-mail]>
> > > Sent: 21 December 2020 17:12
> > >
> > > Hi Lou, WG,
> > >
> > > I find the motivation in the Introduction to be focused on ECA at 
> > > the network devices (with all the talk about issues with 
> > > Centralized network management).
> > >
> > > I see the value of ECA on the controller as well, say a customer 
> > > network controller or an orchestrator can set the ECA on a central 
> > > controller (reference ACTN in TEAS WG). Perhaps you would consider 
> > > adding a sentence to describe this as well. The client-server 
> > > terminology in the rest of the document covers it already.
> > >
> > > And I do see value in this and support adoption.
> > >
> > > <tp>
> > > My take is that the I-D is unclear on what ECA is.
> > >
> > > ECA has been worked on in at least two IETF WG AFAICT.  It cropped 
> > > up in I2RS but as I recall, it was along the lines of 'This is 
> > > ECA'  'No It is not'  'Yes it is' which gave me the impression 
> > > that ECA is not a well-defined, or well-understood, term.
> > >
> > > More recently, I2NSF have produced a YANG capability-data-model 
> > > which is
> > > 55 pages of ECA.  Lacking a definition in this netmod I-D, I am 
> > > unclear what the relationship is between the I2NSF I-D and the 
> > > netmod I-D,
> whether
> > > or not they are using ECA in the same sense.
> > >
> > >
> > Hi Tom,
> > 
> > It usually helps to agree on the problem-space before focusing on 
> > the solution-space.
> > ECA seems like a methodology (ala MVC) more than anything else.
> > The problem statement seems to be that some client tasks need to be
> handled
> > on the
> > server using ECA methodology, instead of on the client.
> > Which tasks? Seems to be any task of arbitrary purpose or complexity.
> > And now the scope is supposed to include controllers (just another
> client),
> > so the problem-stmt
> > is even less clear.
> > 
> > The traditional approach is to pick specific client tasks to move to 
> > the server.
> > The example of detecting and reporting route-flaps has been used.
> > (No ECA example of this complexity has been provided yet).
> > The traditional approach would be to write a route-flap-detection 
> > YANG module with some configuration, monitoring data, and 
> > notification events.
> > 
> > The generalized approach is likely to be extremely complex to 
> > standardize and implement.
> >
> 
> ECA work has a long 20+ year tradition in the IETF and several 
> specifications have been published over the years by various working 
> groups. As far as I can tell, none of them got traction in terms of 
> signifiant deployment of interoperable implementations.
> 
> I would have hoped that the next iteration of ECA work would have 
> started with a deep reflection about why all the previous attempts 
> failed to gain traction and some genuine insights how to design things 
> differently in order to improve the likelihood to have impact.
> 
> /js
> 
> -- 
> Juergen Schoenwaelder           Jacobs University Bremen gGmbH
> Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
> Fax:   [phone]         <https://www.jacobs-university.de/>
> 
> _______________________________________________
> netmod mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/netmod
> 

-- 
Juergen Schoenwaelder           Jacobs University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://www.jacobs-university.de/>

_______________________________________________
netmod mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/netmod