Minutes for NETMOD IETF 97 (Seoul)

Session I
Tuesday, November 15, 2016 13:30-15:30 (2 hours)
Material: https://datatracker.ietf.org/meeting/97/session/netmod
YouTube: https://www.youtube.com/watch?v=m8vm_lX5fXA
Jabber: https://www.ietf.org/jabber/logs/netmod/2016-11-15.html

1.  Intro & Status
Presenters: Chairs 

Comments on: 

<draft-vallin-netmod-alarm-module>`
Kent Watsen: draft-vallin-netmod-alarm-module will be skipped since Martin isn't here.  Would like to hear if WG interested in this draft.
Benoit Claise: some background of the alarm model: 1) I suggest to present it in WG first, and ask whether it make sense. 2) It was shown in hackathon, I required the developer to send it to email list.
Lou Berger: There's a possibility that this draft is more appropriate for a different WG as they have done work on alarms too (CCAMP WG).
Tim Carey: BBF is interested in this work too and having it in Netmod would be better for us.
Lou Berger: We need find right place (WG).
Lou Berger: Which WG is an AD question.
Kent Watsen: How many interested in working on an alarm module?  (some interest)
Lou Berger: We need to ask the other WGs if they are interested in this work.
<\draft-vallin-netmod-alarm-module>


2. draft-ietf-netmod-schema-mount (video offset: 9:45)
Presenter: Ladislav Lhotka
Draft:  http://tools.ietf.org/html/draft-ietf-netmod-schema-mount-03

<is this method complicated?>
Andy Bierman: This seems complicated. Two concerns: 1) the concept of evaluating the when statement, and 2) Normally, all the logic that has to be done for one server; now it different architecture, it has to be done for each subtree.
Ladislav Lhotka: Client does not understand schema-mount; client just use the top model, it does not describe any schema. It is no more complicated than augment. Use only one instance of YANG library.
[Jabber Comment (Balazs Lengyel)] Can we use features to choose between the two ways of reporting mounted data?
Ladislav Lhotka: It is an optional choice.
</is this method complicated?>

<if the primary use cases are impacted?>
Lou Berger: [Slide 13] Not clear which "mount-point" you are referring to in the slide, your presentation mentions it twice. Can you clarify your question? 
Ladislav Lhotka: We are asking if the mount-point extension (which uses anydata) is required
Lou Berger: If you are suggesting to change, what is used by the primary use case (LNE and NI)
Ladislav Lhotka: This will not impact LNE and NI.
Benoit Claise: It's an important issue, we would like to clarify it now.
</if the primary use cased are impacted?>

<two leafs use same term>
Phil Shafer: Two leaves having the same name is confusing
Ladislav Lhotka: Agree
</two leaves use same term>

<use same index values for same module instances>
Dean Bogdonavic: We (LNE/NI) need to be able to use the same list index values (names) for the same module instances under two different NIs.  (This needs to be covered)
Ladislav Lhotka: Here (slides 14) it provides some optional solution to address the open issue (Acee raised)
Lou Berger:  We (LNE/NI authors) have been talking this same issue and have some preliminary thoughts. And we can talk off line.
</use same index values for same module instances>


3. Opstate Design Team -revised datastore (video offset: 43:00)
Presenter: Phil Shafer
Draft: https://tools.ietf.org/html/draft-nmdsdt-netmod-revised-datastores-00

Phil Shafer: how many people have read the draft? (A good number)
<origin attribute and term>
Sue Hares: (slide on section 5.2) Is the "origin attribute" in the intended database?  
Phil Shafer: Intended datastores do not have attributes, it is static.  
Sue Hares: Why called "dynamic" as opposed to something more descriptive?  
Phil Shafer: "yang dynamic" can drive additional format.
Sue Hares: i2rs use "ephemeral", here use "dynamic", it may cause some problem.
Phil Shafer: Don't use the "ephemeral" since it may miss many things.
</origin attribute and term>

<capability>
Dean Bogdonavic: [Slides <intended> #5.1 and <applied> #5.2] what are the capabilities of hardware (box and resources)? 
Phil Shafer: If you pre-configure an interface and it does not exist, then ephemeral interface say that the misconfigured parameters should not show in <applied> configuration.
Dean Bogdonavic: I am thinking of high-level service model, not low-level technology model.
Phil Shafer: It's a use is a particular case.
Dean Bogdonavic: I am thinking QoS, it might be hardware specific models.
Phil Shafer: It's unclear...
</capability>
 
<the relationship between different datastores>
Eric Voit: Do you plan to have relationships about which objects live in which datastores documented?
Phil Shafer: Lifecycle is real-time thing, if something commits, it should be immediately available in <intended>.
Eric Voit: Is it possible to look at metadata and see if it is in intended or applied state? We can take it offline.
</the relationship between different datastores>
  
<regarding open issue slide>
Kent Watsen: Your question if applied would be supported in RESTCONF, but what about the <operational-state> datastore?  
Phil Shafer: <operational-state> datastore would be supported, only question if <applied> datastore needs to be supported in RESTCONF?
Andy Bierman: Do not feel strongly about deprecating <get> operation. Customer uses <subtree filter>, it is easy way to get subtree you wanted.  
Phil Shafer: Deprecating doesn't mean removing, it's means use other rpc to get particular resource. 
Rob Wilton: You can request the restconf to return <operation-state>, but it will not work to request the opstate from running configuration. 
(2:31:34 PM) Martin Björklund: The problem with <get> is that it returns running + state.  Filters are good and should be used in get-config and any new operation we define.
</regarding open issue slide>

<how to connect config data to state data>
[Jabber Comment (Balazs)] : For us the most important part of this work would be: how to connect config data for a managed item (e.g an interface) to the operational state data of the same object. Will you address that?
Phil Shafer: The build of this work is identical. The object in <config> and <operational-state> has same parent edge.
</how to connect config data to state data>

<need to discuss in netconf WG>
Mehmet(netconf chair): process concern regarding which working group this is done in. Netconf's documents are impacted, plan is to present to netconf WG also, and confirm on list.
</need to discuss in netconf WG>

<Opposed to changing RESTCONF>
Andy Bierman: Opposed to changing RESTCONF. Operators do not care if it takes 5 seconds for intended to get applied.
Phil Shafer: The proposal is to add a parameter to allow to get operational-state. 
Kent Watsen: Need to add a bullet item in this slides.  
</Opposed to changing RESTCONF>



4. draft-ietf-netmod-entity (video offset: 1:12:30)
Presenter: Andy Bierman
Draft: https://tools.ietf.org/html/draft-ietf-netmod-entity-01

<Using BBF's augmentations>
Tim Carey: Please consider using BBF's augmentations if you plan to use any augmentations at all. Better, incorporate BBF augmentations directly, because we're doing the same thing.
Andy Bierman: agree. We do not need to duplicate anything.

Benoit Claise: Model catalog in Hackathon needs to be extended to support modules from other SDOs like BBF. And be aware of a yang research web tool which base on catalog to help you retrieve various model information.
</Using BBF's augmentations>

<left to do>
Lou Berger: What is left to do?
Andy Bierman: I don't really know...
</left to do>

<Should model support pre-configuration of hardware components? >
Mahesh Jethanandani: There is an open issue around BBF's proposal for adding pre-configured items. What does it mean to have pre-configured items in the context of NETCONF, and what happens when the pre-configured intended configuration gets applied, specially if the hardware does not exist.
Andy Bierman: We need see the yang model not just yang-tree.
[Jabber Comment Juergen Schoenwaelder]: It would help to have complete proposals and not just fragments sent to the list.
Tim Carey: I will feedback it to BBF and request to send the full yang model to netmod list.
[Jabber Comment Juergen Schoenwaelder]: Confirms that seeing the full model will address his comment.
Benoit Claise: we don't need to wait to have the perfect model, we can ship it and add as needed. Otherwise we risk having perfection that is too late.


#4 draft-ietf-netmod-syslog-model (video offset: 1:25:25)
Presenter: Clyde Wildes (remote participate)
Draft:  http://tools.ietf.org/html/draft-ietf-netmod-syslog-model-11
<WGCL poll>
Lou Berger: How many have read this version? (almost none)
Lou Berger: How many have read any version? (a good number)
Lou Berger: How many think ready for last call? (not a whole lot)
Lou Berger: Any technical reservations?
Dean Bogdonavic: I read this draft before Berlin and think it is ready.
Lou Berger: The chairs believe this draft is ready for last call now that the TLS issue has been resolved. Look for the last call on the list
</WGCL poll>


5. Interface Models update (video offset: 1:31:00)
Presenter: Robert Wilton
Draft:  http://tools.ietf.org/html/draft-ietf-netmod-intf-ext-yang-02
Draft:  http://tools.ietf.org/html/draft-wilton-netmod-intf-vlan-yang-04

<IEEE liaison quesion>
Mikael Abrahamsson: 1) Is there any text on the scope of the IEEE liaison? 2) Does it just cover this draft 
Rob Wilton: 1) Yes; 2) Yes.
Tim Carey: The VLAN draft. What are the nuances related to making the draft informational?
Rob Wilton: VLAN is IEEE domain, and they do not want to standardize it outside of IEEE. Informational is ok.
</IEEE liaison question>
<VLAN draft call for adoption poll>
Lou Berger: The #1 interesting data point is that they are okay with the IETF working on their data plane, but don't want this to be a proposed standard. The "intended status" question will come up again as part of WG last call process.
Lou Berger: How many people think we should work on this VLAN work? (a good number)  
Lou Berger: How many have read this draft? (a reasonable number)
Lou Berger: How many think this is good foundation for the WG? (a good/reasonable number)
Kent Watsen: We will confirm it on the list.
Dean Bogdanovic: Would be good if a silicon merchant vendor would review it. Had issues with Broadcom.
Rob Wilton: Thank you.
</VLAN draft call for adoption poll>


6. draft-liu-netmod-yang-schedule-02 (video offset: 1:44:40)
Presenter: Xufeng Liu
Draft:  http://tools.ietf.org/html/draft-liu-netmod-yang-schedule-02
<different with rpc>
Andy Bierman: Already have time based rpc, it can do same thing. Don't allow a schedule on top of schedule.
There is some promise here, but how this relates to datastores, template exansions, conditional enablement, etc. needs to be worked out.
Xufeng Liu: this different from RPC, because RPCs are not persisted; you can't see them in datastore.
</different with rpc>

<realtime change>
Jason Sterne: Clarification, is a scheduled change the same as if the client had done it in realtime?
Xufeng Liu: correct!
</realtime change>

<similar work in DTN>
Rick Taylor: We have similar work going on in our WG(dtn), and we're in the process to move it to YANG. Let's present it in dtn WG.
</similar work in DTN>

</Ability to modify is needed>
[Jabber (Martin  Björklund)]: What about interactions with locks, access control, validity..? These are details to be worked out. An rpc-based solution is much simpler in this regard.
Tim Carey: Martin likely means how do I know about the authorization of the change at the time it occurs?
Tim Carey: What is the "usage example" slide trying to say? What you want to do on the target?
Xufeng Liu: Yes, it's a target. The schedule can modify the configuration.
Tim Carey: It needs an explicit definition of what you want to do.
Xufeng Liu: OK.
Andy Bierman: Ability to modify (not just add/delete) is needed.  Back to Martin's point, the schedule doesn't guarantee that anything will happen.
Xufeng Liu: Agreed, there are failure cases.
Dean Bogdanovic: Is it just schedule for intent? 
</Ability to modify is needed>

<How many interested in continuing discussing this topic>
Lou Berger: More work to be done, is this an interesting discussion?
Mikael Abrahamsson: As an operator, I'd like to say that this is a useful discussion.
Lou Berger: How many interested in continuing discussing this topic?  (a few/enough)
</How many interested in continuing discussing this topic>

[End of working group session.]




IETF 97 minutes
Session 2
Minutes for NETMOD IETF 97 (Seoul)
THURSDAY, November 17, 2016 11:10-12:10 (1 hour)
Material: https://datatracker.ietf.org/meeting/97/session/netmod
YouTube: https://www.youtube.com/watch?v=Pc9o3QNDrUo
Jabber: https://tools.ietf.org/html/draft-nmdsdt-netmod-revised-datastores

1. Intro & Status
Presenters: Chairs 

2. Opstate Design Team -Revised Datastore Update and Next Steps  (video offset: 5:00)
Presenter: Phil Shafer
Draft: https://tools.ietf.org/html/draft-nmdsdt-netmod-revised-datastores-00

<what shows up in "<applied>" vs. "<operational-state>">
Jason Sterne: Issues on example 2 and 3 slides regarding what shows up in <applied> vs. <operational-state>
Phil Shafer: It's <operational-state> not <applied>
</what shows up in "<applied>" vs. "<operational-state>">

<FRU stand for>
Acee Lindem: What does FRU stand for?
Phil Shafer: It's a field replaceable unit.
<FRU stand for>

<'choice' statements also be a SHOULD>
Rob Wilton: Would 'choice' statements also be a SHOULD, similar to 'when'?
Phil Shafer: Yes. It's a very, very, very should
</'choice' statements also be a SHOULD>
<adoption poll>
Kent Watsen: How many have read the draft? (a good number - including jabber)
Kent Watsen: How many would like to adopt in the WG? (a good number)
Kent Watsen: will confirm on mailing list
</adoption poll>

3. Routing Area DT Update (video offset: 20:00)
Presenter: Dean Bogdanovic
Draft:  http://tools.ietf.org/html/draft-ietf-rtgwg-device-model
Draft:  http://tools.ietf.org/html/draft-ietf-rtgwg-lne-model
Draft:  http://tools.ietf.org/html/draft-ietf-rtgwg-ni-model
Draft:  http://tools.ietf.org/html/draft-rtgyangdt-rtgwg-routing-types
<Action can not instead of rpc>
Andy Bierman: (Slide 5) Action statements are relocatable, it's unlike rpc statements. It can not instead of rpc, it's not work.
</Action can not instead of rpc>

<Schema mount issues>
Ladislav Lhotka: Do you expect that all modules will be available when you construct the data model?  
Dean Bogdonavic: It will be all available, but still have several issues (i.e. how to add new modules into LNE?).
Lou Berger: (To Lada) Do you mean schema mount in general, or this specific use (LNE and NI)?
Ladislav Lhotka: No, it's assumed, whether there is use case like peer mount, uses schema dynamically.
Ladislav Lhotka: (Slide 6) Some confusion about the terminology. Whether the absolute Xpath right for here? 
Dean Bogdonavic: We have to think about the administration right to access the information.  
</Schema mount issues>

<routing types>
Benoit Claise: (Slide 8) A good one, but don't try to make  "types" perfect.  Also, be aware of the model catalog ability to search for typedef. 
Xufeng Liu: What is the scope of a typedef for IP Multicast address.
Benoit Claise: This was just an example. Don't have a specific recommendation
Michael Zitao Wang: The multicast yang type defined in LIME CL module now, we defined it as there was none, but we want to reuse a common type
Acee Lindem: Need to move fast so we're not a bottleneck, please move fast
Michael Zitao Wang: We'd like to contribute the multicast yang type to the routing type yang.
Lou Berger: Note that this work is aimed at the Routing WG not this WG, but we still need to be aware of this work
</routing types>


4. YANG Enterprise Module Namespace (video offset: 40:00)             
Presenter: Xufeng Liu
Draft:  http://tools.ietf.org/html/draft-chen-netmod-enterprise-yang-namespace-03
Draft:  https://tools.ietf.org/html/draft-chen-rdns-urn-07
<Are this for vendor-specific modules? Will affect existing work?>
Mahesh Jethanandani: Are this for vendor-specific modules, or extensions of standard models?
Xufeng Liu: It's work for vendor and if vendor want to augment standard model it also useful.
Mahesh Jethanandani: What happens to existing vendor specific models. Do we have to change?
Kent Watsen: This is limited to new models, right?
Xufeng Liu: Yes.
Mahesh Jethanandani: The applicability of this draft should be specified
Lou Berger: What are you using now and is it compatible with this proposal?
Mahesh Jethanandani: We use URL not use URN for our private models.
<Are this for vendor-specific modules? Will affect existing work?>

<will affect other SDO?>
Tim Carey: BBF is also already started using URNs, but a bit differently -- <company>-yang.
Xufeng Liu: We may have a conflict. The Objective is to ensure no conflict with private modules.
Tim Carey: So issue is for organizations without a standard URN
Xufeng Liu: Correct
<will affect other SDO?>

<optional or mandatory>
Rob Wilton: Is this optional or mandatory.
Kent Watsen: I think it will have to be a recommendation (optional), to replace the "http://" convention
<optional or mandatory>

<useful function?>
Lou Berger: how many think this is a useful function? (a good number)
Phil Shafer: what problem is this solving?
Kent Watsen: It can give a namespace for private modules.
Lou Berger: a deconflicted name space that doesn't look like a URL, and provides a deconflicted urn name space
Phil Shafer: but a URL could actually be useful - I can't imagine anyone using this, but there is maybe just me. 
Mahesh Jethanandani: What's the issue of a broken URL, nothing just a vendor's problem  
</Useful function?>

<adoption poll>
Lou Berger: Repoll (less, but still a reasonable number)
Lou Berger: How many think this is useless?  (slightly less)
Lou Berger: How many think this is a mistake (1)
Kent Watsen: How many have read the draft? (a reasonable number)
Kent Watsen: How many would like to adopt this draft? (a good/reasonable number)
Kent Watsen: We will confirm on this list.
</adoption poll>


5. Implementation experience with IETF ACL Model (video offset: 56:35)
Presenter: Sonal Agarwal 

Dean Bogdonavic: If there's a change in the packet match (leaflist to leaf) there's an opportunity to optimize. 

[End of working group session.]