Minutes for SDNRG at IETF-95
minutes-95-sdnrg-1

SDNRG Minutes by Will(Shucheng) LIU

IETF 95 - SDN Research Group Meeting
10:00-12:30     Monday Morning session I
Room Pacifico A

Agenda

1. Administriva and Agenda Bash - Chair (5 minutes)

2. Existing SDN Research Group Topics

2.1 Cooperating Layered Architecture for SDN (10 minutes) - Carlos J. Bernardos
[CB]

- Differentiate the control functions associated to transport from those
related to services -
https://tools.ietf.org/html/draft-contreras-sdnrg-layered-sdn-04

CB: Ask audience to send comments to mail list.
Dean (not sure for the name, too fast): why is the management and control plane
duplicated in both the transport and the service plane? CB: Want to separate
management and control of services with management and control of transport.
Dean (not sure for the name, too fast): you want to have a completely seperated
management and control plane in transport? CB: Yes. CB:     take the question
offline

Tim Chown on behalf of jabber:
<stewart.bryant> Might you not want to manage them differently? You still might
want to config the lower layer, run OAM etc CB: OK. Point taken. Thanks.

3. Special Session on SDN Intent/Policy

3.1 Intent-based Policy Management (30 minutes)
- A view across SDOs (e.g., ONF, MEF, TMF) and within the IETF (e.g., ANIMA,
SUPA, I2NSF):
   1) the nature of policy
   2) an policy-based management architecture
- Presenter: John Strassner [JS]

Dan Druta: Need to consider limitation of technology layer. e.g. you can do
intent policy on cpu only based on you could detect cpu utilization. express
intent based on , I just make sure that we are not going top down,we need
bottom up. I need to know the airport exists before go to it. JS: Agree. before
you write intent, you need negotiation first. Bert Wijnen: On the slides you
said can intent be updated by devices. Are you meaning can intent updated
devices? JS: I mean device update intent. Sheng Jiang: Anima is not charter to
do intent, but we do realizedd it's important, probably in next charter. 
Intent for anima is not generic, only a subset of broader intent. Bert Wijnen:
Intent is not only apply in sdn, not focus only to sdn. JS: Agree Tim Chown:
Relying the question on jabber <Laurent Ciavaglia> for the mic: The initial
point of discussion was if an ASA could issue / update an intent. Dan King: AI,
machine learning in IRTF might related to this work. Might think about propose
in machine learning rg. JS: need more people working on it.

3.2 NEMO (15 minutes)
- An Intent Oriented Network Programming Language
- Presenter: Bert Wijnen [BW]

JS: Go back slide 6. the language has 3 objects, node, link and flow but there
is no standardization or hierarchy proposed to build them. A subset of node
could be a Ethernet port or a network. How you can build yang model based on
that? BW: The yang model has to be defined as well. Intent has to be mapped to
existing yang models with specified attributes. JS: You start with yang and
then built intent from yang language as oppose to start from intent then
translate to yang? BW: People who do the intent should not have to know about
DM. The people who develop the intent engine need to know DM and translate the
intent into proper attributes in DM. JS: How to do that? the purpose of the
language and the model is to make the generation of yang easier. I don't see
how they connect. BW: We should make policy language easy for end users. People
who write or implement the DM should be of lower priority than the people who
actually specified their intent. Steven Bright: The figure on page 6, two
arrows, one from IM to YANG, the other from DM up to intent description
language. I don't think they're showing the same things flowing. BW: No, at
least in supa they are still thinking to have an IM, based on that people
define YANG DM. We also need device specific DM. In cases if they decide not
have IM that's fine. IM in supa is generalized, like human language like
structure of how you model information, then translate into DM base on such as
YANG, mib, or even something new in the future. Steven Bright: Can you
translate from your IM to your intent? BW: I don't know, maybe you could.
        ALI Karimi: Are you talking about characters of the data, content?  Is
        it the DM the characters of your data or cargo of end users
        transporting from point a to b? yes or no?
BW: Yes.
ALI Karimi: If that's the case, what you should do is putting the type of your
DM, and the characters of every model in your presentation. After that you can
go back to the description to the language and put in on your model. BW: Thank
you. Tim: [14:31:37] <Guy Meador> For the mic: It appears that the IDL you
describe is more of a verbal high-level design language, rather than a policy
expression language. Would you please comment on this? BW: We have to design
some languages, what we need to do is to look at the commands in network today,
try to standardize them.

3.3 Open Mic on SDN Intent/Policy: Thoughts, Findings and Challenges

4. Towards Cost-Effective SDN Controller Solutions (15 minutes)
- A new model for the SDN Controller Placement Problem
- Presenter: Marcelo Santos [MS]

Dan King: Was that the rational for the switch to controller relationship?
Because that is a packet in types of function, where you're doing reactive flow
set-up based on incoming traffic. Maybe in a proactive network or something
with a control plane, you're using controller with some rules setup in advance,
the relationship and latency of the delay propagation between the switch and
the controllers maybe different, the requirement changes. MS: Yes. I agree. In
this model we only considered the delay between switch and controller. Dan:
Interesting topics. Continue to discuss in SDNRG.

5. The Role of the Path Computation Element Centralized Controller in SDN & NFV
(25 minutes) - PCE Protocol as an interface from a Central Network Controller,
South Bound Interfaces, Use Cases and Applications - Presenter: Quintin Zhou
[QZ]

Dan King: pce to talk to control plane. what's the benefit of using pce talks
to control plane. How do you handle the things like availability and
consistency of resource information. it be nice to understand what the
motivation of the architecture. I will take this to mailing list.

6. What Gains for DevOps in Telecom Software-Defined Infrastructure? (12
minutes) - Findings from the Service Provider DevOps activities in the
EU-funded UNIFY project - Presenter: Catalin Meirosu [CM] (name missing):
slides 9, that deep down line is very optimisticly presented there. I wonder
the reason that the decrease is so dramatic. Is it because the infrastractor
scale out accordingly? or purely software faults? CM: Yes. Absolutely . The
fact that we have the capability to of scale out according to a pre-defined set
of rules that are both understand by the operators and vendors of those
functions. This could trigger the line if savings. CM: to audience: if you have
data, you are welcome to share with us.

7. SDN & NFV Co-deployment in Cloud Data Centers (10 minutes)
- Presenter: Rong Gu [RG]
Luay Jalil: (Clarifying)Are most of your problems related to openstack?
RG: yes.

8. BGP Flow Specification
- Do you need BGP Flow specification in SDN/NFV Networks?
- Presenter: Susan Hares [SH]

Dan King: bring this to the list.

9. SDN RG State of the Nation (Remaining minutes)
- Charter, Scope, Existing Individual I-Ds, Research Challenges and Next Steps
Presenter: Daniel & Kohei, All