Computing-Aware Traffic Steering (cats) WG Agenda - IETF 124

When: 2025-11-03 09:30-11:30 Montreal
Co-Chairs: Adrian Farrel, Peng Liu
Secretary: Cheng Li

Logistic

Materials: https://datatracker.ietf.org/meeting/124/session/cats
Note taking: https://notes.ietf.org/notes-ietf-124-cats
Meetecho:
https://meetings.conf.meetecho.com/ietf124/?group=cats&short=cats&item=1

Zulip: https://zulip.ietf.org/#narrow/stream/cats

Disruptions

This meeting was disrupted by multiple issues

• Audio balance that made the chairs' microphone feed inaudible in the
  room
• A power failure that took out the projectors and the router that fed
  the chairs' desk
• A building-wide fire alarm
  The chairs want to thank the attendees and the presenters for
  persevering and achieving a successful meeting.

Notes

1. Agenda Bashing, Introduction, & WG Status (Chairs, 10 min)

[Peng Liu] Two drafts: Use cases and framework drafts are under IPR
call. The WG has started to look into solution related drafts.
[Peng Liu] 2 parts of today's meeting: WG documents presentations and
I-D presentations. Please note that the protocol extensions must be done
in the working group that owns the protocol unless that working group
sends the work to CATS. And presentations here don't imply any support
for a specific protocol extension.

2.1. CATS Use Cases & Requirements Update (5 min)

Goals:

• Update of the progress of addressing issues

• Working Group Last Call
  Presenter: Kehan Yao(remote)
  Reading material:
  https://datatracker.ietf.org/doc/draft-ietf-cats-usecases-requirements

  [Adrian] The last call on this document runs until the 18th, but
  obviously sooner is better for comments.
  [Adrian] 3 contributors still need to reply to the IPR call.
  [Kehan] Will chase them to get the reply ASAP.

2.2. CATS Framework Update (5 min)

Goals:

• Update of the progress of addressing issues
• Working Group Last Call
  Presenter: Cheng Li
  Reading material:
  https://datatracker.ietf.org/doc/draft-ietf-cats-framework
  [Adrian] Does the new update R18 address all the comments?
  [Cheng] More than 90%, still need Adrian's confirmation.
  [Adrian] Will check in this week. Once the comments are addressed,
  the WGLC can be issued. Also, still some contributors need to reply
  to the IPR call, please help to chase them.
  [Cheng] Sure.

2.3. CATS Metrics (10 min)

Goals:

• Status updates since last IETF meeting
• Review of open issues
  Presenter: Jordi Ros-Giralt/Luis M. Contreras
  Reading Material:
  https://datatracker.ietf.org/doc/draft-ietf-cats-metric-definition
  [Greg Mirsky] 1.if the document is standardized, will you
  standardize new metrics?
  [Jordi] L1 metrics and L2 metric will be standardized but not L0
  metrics.
  [Greg] In this case, you should follow the RFC9807. It will be
  helpful to refer to the existing Metrics document/YANG RFCs. The
  metrics will develop very quick,
  [Adrian] This work is not overlap with the work in IPPM.
  [Greg] Agree with you. Good to cooperate with IPPM WG. Liaison
  might be needed.

[Cheng] We should pay more attention to this draft when use case and
framework pass the WGLC. Now we still need more comments and reviews
before it becomes mature. Will work on it after the framework draft.
[Adrian] Agree.

A poll:
Are you OK to change this metrics document to Standards Track?
Yes:11
No:1
No opinion:4

[Erum Welling] Haven't read the latest revision of the draft, so I
said "no" for now. Has the update addressed the comments of CATS
interaction with non-CATS environments?
[Jordi] No, this draft only focus on CATS environment.
[Kehan] I have explained this in the use case drafts. It might be not
related to this draft.

3. Individual drafts

[Adrian] Now we're moving on to individual drafts and around the
solution space and I want to echo the point that Peng made up front,
which is that we're not chartered to do protocol solutions work, but we
are chartered to look at the requirements on protocol solutions and
coordinate with the working groups that do work on protocols.
[Adrian] So, just be careful about assuming that anything will get
developed here.

3.1. Yang Model for CATS (10 min)

Presenter: Quan Xiong
Reading material:
https://datatracker.ietf.org/doc/draft-yl-cats-data-model
[Adrian] The chairs believe that there's something useful here that
the working group should be looking at, so we are we are inclined to
look at adopting this. As we do that, I think there will be two chief
questions for the working group.
(1) Do we need Yang models for this work?
(2) Is this the right starting point assuming that many changes will get
done as the solution space develops?

[Joel] I have problems of both questions. We still have not defined
solutions for CATS now, we cannot tell what kind of information is
needed now. Too early for now. The draft also defines some
forwarding/local behaviors, which is a local matter that hard to be
standardized in a YANG model. This is inappropriate.

[Adrian] We will see whether something can be refined out of this
document before talking about adoption.

3.2. OAM framework and requirements for CATS (10 min)

Presenter: Quan Xiong
Reading material: https://datatracker.ietf.org/doc/draft-fu-cats-oam-fw

[Adrian] I am really pleased to see you looking at the operations and
manageability issues around CATS this early in the cycle, too often it's
left really late. Maybe you should be looking at this to drive the
motivation of the YANG model that you previously talked about.
[Adrian] Let me use this as an opportunity to advertise 5706 bis,
which is currently being polled for adoption in OPSAWG. It talks about
operational requirements on protocols. Maybe you'd like to look at that
as a bit of context.
[Cheng] Suggest to ask for review and comments of this draft on the
list. It has been proposed for a long time, but not enough attention
yet. We need to get the WG consensus to move the draft forward.
[Erum] Is there a mechanism for us to gauge what is the cost of doing
CATS (throughput and management)?
[Quan] Thank you. I think the measurement from the ingress for CATS
border to the egress CATS border is existing technology. But we need
methods to measure the path from an egress CATS node to a service
instance. So we may consider to provide some measuring of the metric for
the CATS service. But this document just provides the OAM requirements,
we have not considered any measurement method yet. Do you have any
suggestions for the measurement methods?
[Erum] Thank you. Was thinking in terms of the high-level decision
whether to use CATS or not. We need some kind of benchmarking of
solutions. Is there already consideration for somehow measuring SLAs
here? What are the SLAs for CATS? I don't have a suggestion at this
time. would have to read the details.

3.3. BGP Extension for 5G Edge Service Metadata (10 min)

Presenter: Linda Dunbar
Reading material:
https://datatracker.ietf.org/doc/draft-ietf-idr-5g-edge-service-metadata

[Adrian] I think we need to look a bit more carefully at the
applicability to CATS and also the non-applicability. For example, what
would happen if CATS was trying to run on a Multi-AS system, and what
would happen if we had a lot of dynamic metrics that we needed to
distribute: would your approach be appropriate? We can also look at the
ways that it would be applicable especially capabilities and things like
that. I think that's a piece of work we can do going forward, and it's
not necessarily something that goes in your draft. It might be that you
write a separate very short draft saying the applicability of the 5G
edge metadata to CATS.

3.4. Distribution of Service Metadata in BGP FlowSpec (10 min)

Presenter: Han Mengyao(local)/Yi Xinxin(remote)
Reading material:
https://datatracker.ietf.org/doc/draft-yi-idr-bgp-fs-edge-service-metadata

[Adrian] Linda's draft gave a good example of doing work in IDR.
However, please also consider the applicability to CATS, though the
draft in IDR may be more general.
[Peng] I want to know what information will be distribute. You can
provide some suggestions to the metric draft.
[Peng] Currently, CATS only focus on single domain. When talk about
large scale deployment, do you mean multiple domains? Please make it
clear.
[Mengyao] OK

3.5. CATS metrics exposure with ALTO (10 min)

Presenter: Jordi Ros-Giralt/Luis M. Contreras
Reading material:
https://datatracker.ietf.org/doc/draft-ymg-opsawg-service-deployment-with-alto-cats

[Adrian] Is this document totally focus on CATS? Or does it have wider
applicability?
[Luis] Just CATS
[Adrian] OK. ALTO currently has no home in the IETF. I see you are on
the agenda for OPSAWG. I will check with the ADs and chairs to see where
your work belongs.

3.6. DNS based service discovery for CATS (10 min)

Presenter: Xiang Liu(Remote)
Reading material:
https://datatracker.ietf.org/doc/draft-liu-cats-dns-service-discovery

** Reordered to after 3.8
[Adrian] CATS is not going to do DNS work. We can discuss here, but
please find a good home for this work. You might need to talk with Jim
Mozley (Infoblock), to see how to move forward. He has work on AI agent
and service discovery presented in Dispatch parallel to this session. A
BOF may be needed. Will drop you an email about this.

3.7. CATS Reference Model for AI-Agent Communication Network (10 min)

Presenter: Tianji Jiang
Reading material:
https://datatracker.ietf.org/doc/draft-jiang-cats-reference-acn

[Linda] Do you have a summary of the key differences between ACN and
today's network?
[Tinaji] It is just a global network framework for communication among
AI agents. Some requirements, like on-demand communication. The
realization model also needed interactions and collaborations security
control. At this moment, there is no specific requirement on how KPI
metrics will be achieved these are still being explored.

[Cheng] Any new metric need to be standardized?
[Tianji] For CATS it's an objective things. For AI agent, if it is
instantiated as a virtual format, it will have a different parameters,
but it does depend on the AI models: all kinds of things such as the
number of parameters. So I consider this one: the AI agent is going to
have objective and might have subjective metrics. But the subjective
ones do not need to be considered. So I see a need to consider extension
for the metrics part.

[Daniel Huang] I read the white paper of Agent Communication Network.
I believe CATS framework will help ACN in terms of AI service traffic.
[Daniel] I'm confused: you mentioned AI agent metric is application
metric. My understanding is that the AI agent metrics could be aligned
between CATS framework and the ACN framework, so would suggest the CATS
metric remains as it is in Kehan's document, and you can make other
sections to show the alignment between the ACN framework and the CATS
framework, and what needs to be done to fill the gap.
[Tianji] Good suggestion

[Luis] I was wondering if the data that support the processing of the
agent is also considered here? It's more kind of subjective. Try to make
it objective things so it's like a compute metric / network metric that
kind of a very objective things and then we consider the data traffic.
[Tianjin] We try to standardize the object metrics.

[Adrian] At the moment, CATS is not chartered to look at AI agents. It
is an interesting problem and it is potentially an extension of CATS,
but we need to be careful. So I would suggest keep doing the work, keep
talking to all the other AI agent groups. There's a lot of interest.

3.8. Applicability of CATS Framework (10 min)

Presenter: Tianji Jiang
Reading material:
https://datatracker.ietf.org/doc/draft-jlmcyp-cats-applicability
[Chuanyang] Page 6. Do you think the AI agent is a kind of the service
instance or is the different things?
[Tinaji] It's different from the service in the context of CATS right
now, because the service in CATS is more like a service instance,
associated with some service metrics.

4. Open Discussion (Chairs and All, 10 mins)