DetNet Minutes for December Virtual Interim 2022/01

WG ICS: https://datatracker.ietf.org/meeting/upcoming.ics?filters=detnet

Datatracker: https://datatracker.ietf.org/group/detnet/about/

Monday, December 12, 2022

13:00 - 15:00 (UTC)
https://www.timeanddate.com/worldclock/fixedtime.html?msg=DetNet+Interim&iso=20221212T13&p1=1440&ah=2

Materials:
https://datatracker.ietf.org/meeting/interim-2022-detnet-01/session/detnet

Note taking:
https://notes.ietf.org/notes-ietf-interim-2022-detnet-01-detnet?both
Meetecho:
https://meetings.conf.meetecho.com/interim/?short=7d4a462b-ffbb-47b0-aaac-6b511e2817f7

Jabber Logs: http://jabber.ietf.org/logs/detnet
Session ICS:
https://datatracker.ietf.org/meeting/interim-2022-detnet-01/session/30066.ics

YouTube: https://www.youtube.com/watch?v=89Fx9jaG_NQ

Slot Start Time (UTC) Duration Information

#1 13:00 15 min Title: Intro, WG Status, Charter, Draft Status

Presenter: Chairs

Lou Berger presents.

#2 13:15 10 min Title: Requirements of large-scale deterministic network

Presenter: Peng Liu

Draft: https://datatracker.ietf.org/doc/html/draft-ietf-detnet-scaling-requirements-00

(13:07)
Peng Liu presents.
David Black:suggest to distinguish between tolerating fluctuation and recovery from (reduction of) introduced fluctuation
Quan Xiong: Gap analysis document has submitted and hope to give more clarification
Greg Mirsky:Do you see it is possible to quantify large latency? Bounds on what latency can be tolerated? 
Lou Berger: The request is to quantify in the document what is meant by large latency.
David Black: Unbounded jitter is even more difficult than unbounded latency - please include that in the definition.
Toerless Eckert: Would some examples of (industrial) application latency help?
Greg Mirsky: We need Link/Path propagation latency and jitter information in the document (not application info).

#3 13:25 10 min Title: DetNet Enhanced Data Plane Requirements

Presenter: Xuesong Geng

Draft: https://datatracker.ietf.org/doc/html/draft-yzz-detnet-enhanced-data-plane-01

(13:22)
Xuesong Geng presents.

Janos Farkas: Please work with authors of requirements document/mailing list to incorporate requirements into the newly adopted WG document.

Lou Berger: Enhanced data plane document is currently called "scaling", please work with the authors of that doc, come up with proposed additions, discuss on list. Authors have a lot of flexibility on how/when to bring in proposed text.

David Black: Seconds that remark from Lou. Design team is also in his opinion expected to work from revised version of that document.

David Black: Would like to see WG discussion about whether or not (re-) use DSCP. This is about how DetNet will relate to DiffServ architecture.

Lou Berger: This slot is for requirements. You have another slot for solution details. Please bring solution in the other slot.

Yizhou Li: Open to taking enhancments into existing WG scaling requirments document. Scaling is considered one of the root causes at the beginning. It would be great to take other considerations into the requirement document as well.

Xuesong Geng: Willing to bring requriements to the WG document.

Peng Liu: We have changed the name from large scale to scaling. It gives more chances to cover more requirements.

#4 13:35 15 min Title: An Information Model for DetNet packet processing metadata (and beyond)

Presenter: Toerless Eckert

Draft:

(13:40)
Toerless Eckert presents.

David Black: You are discussing two things. Taxonomy is very useful. Encoding will be a fairly long discussion.
    Should start with taxonomy and decide what needs to be communicated before discussion of how many bits will be spent on each information element.
Toerless Eckert: Want to start with semantic elements of queueing mechanisms. Need to collect this info, with help from authors.

Xuesong Geng: Considering David's question, we are also trying to introduce metadata for queueing, so define two types of bounded information, 1st is time stamp, which will limit number of bits; 2nd is Resource ID, so classification of queuing mechanisms will help.

Yizhou Li: Have various metadata. It would be good to identify what is common. Identify fixed fields of MD, which are mechanism-specific md. There are fixed vs extensible MD fields. Second the opinion that MD formatting is important, but neeed to reach consensus that a specific queueing mechanism will be useful here.

Toerless Eckert: There is a limit on how much we can carry in packets. The more we want, the more pushback we will get. So consider the difficulty of our asks; need to keep this in mind.

Balázs Varga: I think simplification is a very good direction. We should limit the information we add to DeNet packets for helping bounded latency related scheduling along the forwarding path. Whether it is one or two parameters, or only just class information is up for discussion.

#5 13:50 10 min Title: DetNet Enhanced Data Plane

Presenter: Li Zhang / Xuesong Geng

Draft: https://datatracker.ietf.org/doc/html/draft-yzz-detnet-enhanced-data-plane-01

(13:54)
Li Zhang presents.

Daniel Huang: Confused about suggestion of draft that end to end budget will be encapsulated in data plane. Why is it necessary to know ambient delay budget? End to end service requirements can be part of control plane. Don't need it as dataplane MD for DetNet.

Li Zhang: End to end delay is useful for when it would require too much time to forward the packet. So use timer to forward it. Choose a required queue based on delay left.

Xuesong Geng: I can add some info for Daniel's question. For E2E may use other mechanisms not used so far in IETF. How to use at intermediate/transit node. If have E2E budget then amount used in node can be subtracted so keeping track of time budget. But proposal is not to carry some specific MD, is more like Toerless' idea of common mechanisms between queueing mechanisms. Need to identify which MD items these are, then can discuss in WG.

Toerless Eckert: Some proposals that include a deadline, even Yakov would say they are based on heuristics, not calculated. So may be useful to share packet headers but may not be deterministic.

#6 14:00 5 min Title: DetNet Data Plane - Tagged Cyclic Queuing and Forwarding (TCQF) for bounded latency with low jitter in large scale DetNets

Presenter: Toerless Eckert

Draft: https://datatracker.ietf.org/doc/html/draft-eckert-detnet-tcqf-01

(14:06)
Toerless Eckert presents.

David Black: About use of DSCP, just because you can, doesn't mean you should. Benefit of this document is that a small number of bits is needed. Need to describe the number of bits needed; presumably they could fit into DSCP, but need to discuss. Focus on number of bits needed, not where to put them.

Toerless Eckert: Yes but takes long time to get things changed at HW level, want to reuse what we can from existing packet header.

David Black: Separate what needs to be done from how to do it; TCQF gets a lot of leverage out of a few bits. Correct about implementation constraints, but it is an architectural decision, e.g. to use DSCP was decided against for initial DetNet, now revisiting that. This is architectural, not a consequence of implementation constraints on where bits can be carried.

Xuesong Geng: Question of whether DSCP is appropriate for use in DetNet - consider its relationship to Diffserv architecture.

Toerless Eckert: Used in history, e.g. A41, A42, so wouldn't be first time to consider more than one DSCP for a flow. So don't have architectural reason for why DSCP would not be appropriate for use here, but could consider the matter.

Janos Farkas: This is good input for the design team, so we should follow up with them. 
Lou Berger: Agree.

#7 14:05 10 min Title: Towards Large-Scale Deterministic IP Networks: variant of CQF

Presenter: Yizhou Li

Drafts: https://datatracker.ietf.org/doc/html/draft-yizhou-detnet-ipv6-options-for-cqf-variant-01

https://datatracker.ietf.org/doc/html/draft-eckert-detnet-tcqf-01
(14:13)
Yizhou Li presents.

Lou Berger: Leave discussion til after next presentation.

#8 14:15 10 min Title: More technical details on the scheduling methods proposed in the ADN draft

Presenter: Jinoo Joung

Drafts: https://datatracker.ietf.org/doc/html/draft-joung-detnet-asynch-detnet-framework-01

(14:24)
Jinoo Joung presents.

(No discussion)

#9 14:25 10 min Title: Deadline Based Deterministic Forwarding

Presenter: Shaofu Peng

Draft: https://datatracker.ietf.org/doc/html/draft-peng-detnet-deadline-based-forwarding-03

(14:36)
Shaofu Peng presents.

David Black: At slide 8 deployment considerations: Are you suggesting that there be 2 DetNet services, one is bounded latency, "in-time", vs "on-time" that delivers at precise deadline? 
Shaofu Peng: Yes. 
David Black: How does that interact with the current DetNet architecture that focuses solely on bounded latency? This is a question to follow up on later. Also I don't understand your "Cost" definition: Deadline-based scheduling requires some notion of time - what is the assumption on time synch here? 
Shaofu Peng: For DetNet we would not depend on time synch. It is based on offset of time, just using local time. Can discuss on list. 
David Black: Something doesn't make sense about this picture. 
Shaofu Peng: We can clarify.

#10 14:35 10 min Title: The queuing based enhancement in data plane and control plane

Presenter: Quan Xiong

Drafts:

https://datatracker.ietf.org/doc/html/draft-xiong-detnet-6man-queuing-option-02

https://datatracker.ietf.org/doc/html/draft-sx-detnet-mpls-queue-03
(14:46)
Quan Xiong presents.

#11 14:45 15 min Open discussion

(14:55)

Lou Berger: We expect the main discussion to be in the design team, which is open for all to participate.

Toerless Eckert: On the ADN draft: Scalability is related to CQF, so if we have 100 ingress nodes, we have 100x100 = 100000 flows to aggregate. 
This requires per-hop per-flow steering state - is this scalable? Need more input from WG on this.

Jinoo Joung: Yes, correct about scalability isue with ATS, need to maintain per flow state, so FAIR and PFAR is mitigation for this, particularly port flowing regulator doesn't have that issue. Also SCSCORE (?) . Will discuss in design team.

Lou Berger: There is some similarity in the drafts. I encourage authors to work in author groups informally, while waiting for design team. Can use IETF resources - you can contact the chairs alias to set up such meetings.

Lou Berger: There is a lot of energy on this topic, we appreciate that in a contribution-driven organization.

Janos: Thanks and happy holidays.

Lou Berger: See you in upcoming meetings. Thank you all.

15:00 Adjourn