SUPA: 7/22/2015 Dan Romascanu and Tina Tsou chairing Based on notes taken by Susan Harris and Andrew Veitsch 0. Agenda Bash Discussion: none Dan providing introductory comments Description slide presented Simplified Use of Policy Abstractions Define a generic policy information model GPIM SUPA defines a vendor neutral mechanism to express management policies. It is a working group forming BoF. * Robert Sparks: Your charter indicates you are going to have a new network management interface. Is this correct? * Dan: Our definition of this work takes high-level policy and outputs management snippets. We do not expect to specify how the logic works, only how it is activated. 1. Goals, Framework, Examples (Maxim Klyus) Discussion: * Bert: I do not see why one is declarative and one is imperative. * Maximum: The first one does not have a condition. * Bert: The first one says ?if the request is not source? * Dan: I think a different example would better describe this point? * Robert: Many of the examples miss that the same words - Provide an example to show the difference. * John Strassner: Perhaps a clearer example for declarative - where you ask to compute a distributed transitive closure. In the declarative, iterate over all elements in the subnet. I do not know which elements I am working on. o Does this help? SUPA Examples for VPC (Dacheng Zhang) Discussion: No comments SUPA : Load Balance when Link Load is too high (Oscar Gonzalez) Discussion : * Benoit: How do you plan to use SUPA model with intent model and declarative? How will you use it? * Oscar: I would then use protocol with my systems. I would like to know what requests I would use to PCE. A general model will make it easier for me because I do not have to create my own. Without a standard model, I need to spend a lot more money iterating to find the right model. At least I can express a set of parameters at a high level. * Benoit: I do not need to expand the models for the services. Will they include the lifespan or lifecycle of the planned policy? * Oscar: Had not considered. * Max: The Company implements all this for the customer. We would like to make it * Ying Cheng: What are the problem management systems? * Dan: This is not clarification questions. * Dean Bogdanovic: Looking this, how differs from L3SM? * Dan: next point in the agenda may clarify 2. Relationship to other WGs (Sue Hares) Discussion: * Dean Bogdanovic: differentiation and the SUPA work and the other groups - is not that big - not a clarification question ? Dan R. moved it to later * Jabber question, Laurent: gap analysis relative to the ANIMA activity? Intent is being used there - similar to what is being described here. * Anima chair - Sheng Jiang: intent not yet in the charter. Thinks it is quite different - no overlap * John explained this with a motivating example. 3. Information Model Overview (John Strassner) Discussion * Jamal Salim: You will have a difficult time with approving the information model in the IETF. * Dan: I believe you are incorrect. We checked with the OPS ADs and IESG who said if there are good reasons to develop an information model the WG may decide to do it, and it will be approved 4. Data Model Status (Tina on behalf of the team, Andy Bierman was sick and could not attend the meeting, 10 minutes) Discussion * Andy Veitch: is the dependency or association of a policy with one or more (plus versions) of network, topology etc. models in the policy data model? * Dan: I believe that it should be but it?s not all clear by now. 5. Charter presentation (Juergen Scoenwaelder, 15 minutes) Discussion: [Slide 1]: * Robert: The charter says the information model will be translated into data model(s) - is this group going to do that translation?? * Juergen: The group says it will translate the information model into the data model. * Jamal: If there is an information model * Liu: The translation is the clarification on what it will say. * Kareeti: The WG will define the information model to the data model translation or the engine performing such operations? * Juergen: You have information model, Yang model, and the engine is translating it to the other work. [Slide 2] * Adrian: SUPA defines this technology - should give the specific data model and technology model. [Slide 3] * Adrian: SUPA will be designed to work together with the network service models? SUPA is a working group ? a technology? needs clarification. * Benoit: The previous slide it will not focus on the service. Why the duplication? * Juergen: We had a long discussion to make sure the SUPA was unique. This text is a leftover from that discussion. * Dean: There will be an information model will be translated into a data model? Kareeti question indicated that the information model into a data model. * Michael Scharf: I totally understand why you have a service model twice. I think that we have a service model. * Dave Dolson: Can you give me an example that is not? * Juergen: An ACL provides policy on a network interface. The SUPA is trying to write on top of the ACL. ACL is bonding to the service model. * Dave Dolson: It may produce a network element that is in policy. This is unclear. * Pedro: You can reduce this charter into few sentences. I encourage you to provide a few sentences. * Sam: If those are not in scope of other working groups? Will these be in scope? * Dan: yes * Dave Dolson: I see nothing that has first order logic. * John: Declarative logic is first order logic. The market calls it intent, and I call it declarative. * Dacheng Zhang: The use case output is not included in the charter. * Juergen: The charter does not say we are going to create use cases. The availability in the charter can have some examples on how to apply * Benoit: I want to see the end goal. Do you intend to do the information model and the data model? * Dan: Yes * Benoit: In the end you need to automate the data model - I do not picture what you have for intent model? * Dan: The model will have expressions plus the context of the models. We will have a filter context. This is how I imagine it. * Benoit (to Juergen): You are a yang expert - do you believe this work is achievable. * Juergen - Yes * John: I do not know how to accomplish the declarative in yang. Our approach is to invent this * Michael Scharf: The SUPA will describe configuration models. * Dan: The devices will * Scharf: It is unclear for the charter. I have doubts has this will appear. * Juergen: what phrase do you mean? * Michael Scharf: I do not understand the scope of #4. * Juergen: My understanding is the intent. The other deliverable are technology documents, and the fourth is an unclear technology document. * Dan: It is a statement that says what it is about, and what is not in scope. * Eric Voit: Is there any plans for revocation of intent? Is there a condition that will be included in the charter? * John Strassner: There are two ways to prepare this point. For intent, I could implement positive and negative logic statements * Eric: There are ongoing conditions in the network. * John Strassner: A query that looks for queries in the network, and provides a clarification. * Jamal: Is an example of the choice of data models? * Dan: NOT tied to Yang only, but Yang is the example of the data models. * Jamal: We do not provide generic models any more, but SUPA. * Dan: Warning: we have 10 minutes left. * Uni: For the work. need more information to approve or not? * Dan: Is it experimental or actual? We had actual. * Uni: I think we should go forward to discussion. * Dan: We will skip forward 6. Discussion on whether the progress assessment implies going forward changes to the charter (all, 30 minutes) Discussion: * Cheng Ying (China Unicom): Data center migration and diagram. Manual migration is very difficult. We think the policy that this proposes is useful. * Linhui Sun (Tsinghua University): Every vendor has its own model for data models. It will provide a generic model that will guide. * Chongfeng Xie (China Telecom): I will state the person that has stated we had a need for the data models. I would like to note that we need the automation work that is being work. I like declarative model for the OAM. Isuggest SUPA team to design solution to address this problem * Ravi: Are we trying to leverage to provide existing work from Datalog? * Dan: Are you discussing how the information model is expressed? * John: Compared to Datalog, we are providing a simpler version of logic (called propositional logic), and a more expressive version (called first order logic).? We could add pure Datalog (and/or OpenStack's extensions to it) if desired. * Ravi: You could perform the preference. * [XXX]: The I2NSF is working on the security layer interface. Is this WG going to handle security? * Dan: I have the privilege to chair both I2NSF and the SUPA. If both are approved, the I2NSF will be one of the output. * Dean: SUPA defines about the higher policy data model to the SUPA yang model. Into the policy data model you will translate to the service data model? * Juergen: We do not translate into service model? * Dean: We may point to a service data models. You go from service to a device. * Dan: Sometimes you do. * Dean: You will define. * Juergen: I believe the service model may trigger the policies. Does this make sense? * Dean: I do not understand how you differentiate this from other WGs. * Kareeti: I am a huge fan of Declarative policy. I think we are early. It is half-way declarative model. One of the things that I took away, is that you are above it. It is not my place, but I suggest this may be at the IRTF. We are not ready to put this in the declarative. I am seeing more head in the clouds than feet in the clouds. * Adrian: I thought we skimmed over the slide with the statement that said - the policies for any WG or Data models from this point that are not impeded. I am not clear in the value of information models. The progression is normally to start an information models and go to data model. Pretty soon the data model 7. Questions to the audience (chairs, 20 minutes) Discussion: * Is the problem statement clear, solvable and useful? o 50% /50% - solvable and not-solvable * Is the charter useful? o 60% (yes)/ 40% (no) * If such technology model (information model and data model), are you interested in deploying it? o Who is an operator ? 12 o Who operators would interested in deploy it? 7 * People who are interested in write code for data models? data models? 15 * People who are interested in writing documents? 15 * People who are interested to review documents? More than 20 Wrap-up: * Benoit: Very balanced hums in the room - confusion on generic objects - and more work - wants to understand what the problem is - wants to discuss with the IESG etc. I will work with the folks to clarify the problem issue in the charter that we are not understanding. * Dan: Thank you for the hard work.