Re: [Actn] ACTN charter v1.0

["King, Daniel" <[e-mail]>]

Hi Young and Richard,

Just to be clear then, the definition for an “overlay network” is as follows:

   Overlay Network: An overlay network runs independently on top of
   another network. The nodes in the overlay network may considered
   connected by virtual or logical links, each of which corresponds to
   a path. Virtual Extensible LAN (VXLAN) and Network Virtualization
   Using Generic Routing Encapsulation (NVGRE) are considered overlay
   network technologies.

Also, the definition of “network virtualization” below might be better phrased as “network virtualization interface”:

      Network virtualization *interface*, refers to allowing the customers to
      utilize a certain network resources as if they own them and thus
      allows them to control their allocated resources in a way most
      optimal with higher layer or application processes. This customer
      control facilitates the introduction of new applications (on top
      of available services) as the customers are given programmable
      interfaces to create, modify, and delete their virtual network
      services.

Then we could update the definition of “network virtualization” to:

   Network virtualization is the process of virtualizing network
   resources and facilitating the network to be programed, administered and
   deployed or released, independently of the underlay transport mechanism,
   hardware or software, and operations, administration and management
   method.

Of course, just to make things more confusing, if you talk about network virtualization to a data center provider then it will probably include virtual network functions (firewall, WAN accelerator, et al.). So we may also want to make that clear for the NFV folks ;-)

Br, Dan.

From: Leeyoung [mailto:[e-mail]]
Sent: 17 September 2014 19:41
To: King, Daniel; Y. Richard Yang
Cc: [e-mail]; Daniele Ceccarelli; Dhruv Dhody
Subject: RE: [Actn] ACTN charter v1.0

Hi Richard,

Thanks for great comments. I think Dan provided good responses to your comment. Please see below for further clarifications on some of your comments.

- The current version appears to trying to do both (1) imposing an architecture and (2) leaving architecture to the architecture document. For example, I see multiple types of controllers (physical controller for a domain, coordinator for physical controllers, tenant controller for a tenant, ...). I also already see how they may interact (e.g., the hierarchy paragraph). Some email discussion start to touch on this issue. My preference is that it is better to leave the architecture as much as possible to the architecture document, to avoid confusion, because there are so many possibilities (e.g., a controller might be represented by two functions, one for abstraction, one for control), unless the mailing list can reach some agreement soon.

YOUNG>> Multiple types of controllers (as well as their functions) discussed in the charter are based on use-cases of various operators. When discussing info/data models and interfaces and protocol extensions, we ought to have entities of main interest. This gives contexts of relevant entities at a minimum level.

- The charter uses the word "virtualized" quite a few times. I found that they appear to have different meanings. For example, in this sentence, "Well-defined use cases from operators perspective with clearly stated need for transport network virtualization", does "virtualization" mean abstraction? For this sentence, "a centralized virtual network operation", I cannot fully understand the meaning of the word "virtual". For this sentence, "Multi-tenant support to allow virtual network information query", it appears that virtual means an overlay network, right?

YOUNG>> Virtualization is to be differentiated from abstraction. Abstraction is a way of representing network resources and topology (hiding details from physical layer). Virtualization allows operators to make use of abstraction to simplify its network operation as well as to offer services to their customers which is known as network slicing. Here’s an attempt to define what is network virtualization from the problem statement draft https://datatracker.ietf.org/doc/draft-leeking-actn-problem-statement/ :

o Network Virtualization:

      Network virtualization, refers to allowing the customers to
      utilize a certain network resources as if they own them and thus
      allows them to control their allocated resources in a way most
      optimal with higher layer or application processes. This customer
      control facilitates the introduction of new applications (on top
      of available services) as the customers are given programmable
      interfaces to create, modify, and delete their virtual network
      services.

Overlay network is a form of virtual network. Overlay network in the context of transport networks arise in a multi-layer context where the packet layer is a client to the server transport layer.  Packet layer is given a topology comprised of virtual links and nodes (typically in a tunnel view) from transport layer.

Thanks,
Young


From: King, Daniel [mailto:[e-mail]]
Sent: Monday, September 15, 2014 1:45 PM
To: Y. Richard Yang; Leeyoung
Cc: [e-mail]<mailto:[e-mail]>; Daniele Ceccarelli; Dhruv Dhody
Subject: RE: [Actn] ACTN charter v1.0

Hi Richard,

Good comments. As with most complex proposals, having clear definitions, terminology, and conceptual view is fundamental to understanding the system and scope of work. My specific comments are inline (“DK>>”):

Br, Dan.

From: ACTN [mailto:[e-mail]] On Behalf Of Y. Richard Yang
Sent: 12 September 2014 23:47
To: Leeyoung
Cc: [e-mail]<mailto:[e-mail]>; Daniele Ceccarelli; Dhruv Dhody
Subject: Re: [Actn] ACTN charter v1.0

Hi Young, all,

This sure is an active list and it is a bit hard to jump to the discussion! I read charter v1.1 and the discussions followed on the changes. Here are some early comments:

- The current version appears to trying to do both (1) imposing an architecture and (2) leaving architecture to the architecture document. For example, I see multiple types of controllers (physical controller for a domain, coordinator for physical controllers, tenant controller for a tenant, ...). I also already see how they may interact (e.g., the hierarchy paragraph). Some email discussion start to touch on this issue. My preference is that it is better to leave the architecture as much as possible to the architecture document, to avoid confusion, because there are so many possibilities (e.g., a controller might be represented by two functions, one for abstraction, one for control), unless the mailing list can reach some agreement soon.

DK>> You make a good point, I think we became blinkered slightly and have maybe assumed a number of ACTN architectural principles. We can look to be less prescriptive in the charter text when we issue the next version.

- In the document, the word "abstraction" is a keyword, and it is different from "control", which is the other keyword. I am fine with keeping the meaning of abstraction somehow vague, but it may become confusing when there is too much overloading.

DK>> Ok. We can also check the charter against the use cases and problem statement to ensure constancy across I-Ds.

For example, according to this sentence, "the creation of a virtualized environment allowing operators to view and control multi-subnet, multi-technology, multi-vendor domain networks", abstraction is similar to "view". Then, the following sentence "Abstraction of transport networks also allows operators to consolidate their network services into multi-tenant virtual transport networks." appears to assign abstraction a much stronger meaning. I would suggest revision to keep the meaning of the keywords simple.

DK>> Agree, also this does appear to be a solution statement than a requirement or objective.

- The charter uses the word "virtualized" quite a few times. I found that they appear to have different meanings. For example, in this sentence, "Well-defined use cases from operators perspective with clearly stated need for transport network virtualization", does "virtualization" mean abstraction? For this sentence, "a centralized virtual network operation", I cannot fully understand the meaning of the word "virtual". For this sentence, "Multi-tenant support to allow virtual network information query", it appears that virtual means an overlay network, right?

DK>> Yes, we need to be careful with consistency and usage of terms. We must avoid interchanging terminology and should be clear on how we use “virtualization”, “abstraction” and” multi-tenant”. In an early version of the ACTN problem statement we included a survey of existing technologies. To help set context and what we mean by network overlay and network virtualization, it might be worth discussing these technologies again (e.g., VXLAN, et al.), and what is required for a network overlay versus network virtualization.

- Increasingly, "abstraction" in networking appears to mean (declarative) information model/data models, and this is what I read from this document as well. But abstraction can mean a lot more.

DK>> You made the same point at the BoF in Toronto and it is still relevant. A protocol/forwarding specific data model, and info model, is not a trivial matter. Other WG initiatives are investigating and developing solutions and as you suggest, we should document what would be required from an ACTN perspective and feed into those initiatives. We can probably state this more clearly in the charter text.

Br, Dan.

On Fri, Sep 12, 2014 at 11:07 AM, Leeyoung <[e-mail]<mailto:[e-mail]>> wrote:
Hi Daniele,

Thanks for providing your good comments here. Some of your comment (1) is for charter discussion while the others (2 & 3) are architecture discussion. As a few folks are already working on the architecture document, we may start a new email thread on architecture discussion while developing the draft.

Please see inline for my response.

Regards,
Young

From: Daniele Ceccarelli [mailto:[e-mail]<mailto:[e-mail]>]
Sent: Friday, September 12, 2014 9:27 AM
To: Dhruv Dhody; Leeyoung; [e-mail]<mailto:[e-mail]>
Subject: RE: ACTN charter v1.0

Dhruv, Young,

Excellent progresses!
Let me just add some notes/questions.
I’m collecting them here and not inline as the reading of the whole thread is becoming pretty complex.


1.      On centralized signaling: We need to clearly state what centralized signaling means. NMS based is one of the possibilities but it’s not limited to NMS. A network implementing e.g. Open Flow or any other similar protocol should be considered as well IMHO. Here I’m not sure whether we can speak about centralized signaling in both cases or not. In the case of NMS we use the management plane to dynamically provision connectivity (whatever protocol is used). In the case of e.g. Open Flow can we speak about management plane or is something else? Please note that I speak about OF as it’s the only example that comes to my mind but any other possibility can/must be considered.

YOUNG>> The intention here was giving some overview of the tools that current transport networks have. Of course thing are quickly changing with other initiatives like OF and even PCE based signaling in a research circle that can be viewed as centralized signaling. Perhaps good compromise would be putting NMS based as an example. How about restating like:

Allow for distributed signaling or centralized signaling (e.g., an NMS-based, etc.) for set up….”



2.      On multidomain issues: One question here: How do we want to manage multi-domain? I see two ways:

a.      A coordinator of virtual network controllers – i.e. 1:1 relationship between virtual network controller and physical network controller plus a coordinator of virtual network controllers on top

b.      A coordinator of physical network controllers – i.e. a coordinator of physical network controllers (1:N relationship between coordinator and PNCs) with a virtual network controller on top? (1:1 relationship between PNC coordinator and the VNC)

Maybe we should consider both architectures, maybe pick one.

YOUNG>> These options seem to be detailed architecture that the upcoming architecture document should address in detail. My personal opinion is that both architectures need to be considered as a starting point. There may be other variations. Then we may narrow the scope as an initial baseline architecture.


3.      Again on multiodomain: maybe we could improve the architecture draft saying what is centralized and what is distributed. E.g. VPN prefixes exchange could be distributed (controllers speaking MP-BGP with each other) while provisioning could be hierarchical (i.e. the coordinator asks controller A to provision a path with domain A and asks controller B to provide a path within B so to have an end to end path crossing domains A and B)…H-PCE like.

YOUNG>> Good point. I agree that the architecture draft should discuss these dichotomy of control regimes (e.g., centralized vs. distributed). One question I have: are you assuming the coordinator also speaks MP-BGP with lower level PN controllers? If not, would the coordinator need to collect the VPN prefixes using the interface between the coordinator and each PNC (that speaks MP-BGP)? Then can this be viewed as a centralized component?

BR
Daniele

From: ACTN [mailto:[e-mail]<mailto:[e-mail]>] On Behalf Of Dhruv Dhody
Sent: venerdì 12 settembre 2014 04:33
To: Leeyoung; [e-mail]<mailto:[e-mail]>
Subject: Re: [Actn] ACTN charter v1.0

Hi Young,

Please see inline..

From: Dhruv Dhody
Sent: Wednesday, September 10, 2014 10:48 PM
To: Leeyoung; Leeyoung; [e-mail]<mailto:[e-mail]>
Subject: RE: ACTN charter v1.0

Hi Young,

Please find some comments on the proposed charter.

* Do we need some text in charter also to specify what we consider as Transport networks?

YOUNG>> That may clarify the scope. Do you have any suggestion?  One definition from the Problem Statement draft (https://datatracker.ietf.org/doc/draft-leeking-actn-problem-statement/) is as follows. Would this be good enough or need more work?

Transport networks are defined as network infrastructure that
   provides connectivity and bandwidth for customer services. They
   are characterized by their ability to support server layer
   provisioning and traffic engineering for client layer services,
   such that resource guarantees may be provided to their customers.
   Transport networks in this document refer to a set of different
   type of connection-oriented networks, which include Connection-
   Oriented Circuit Switched (CO-CS) networks and Connection-
   Oriented Packet Switched (CO-PS) networks. This implies that at
   least the following transport networks are in scope of the
   discussion of this draft: Layer 1 (L1) optical networks (e.g.,
   Optical Transport Networks (OTN) and Wavelength Switched Optical
   Networks (WSON)), MPLS-TP, MPLS-TE, as well as other emerging
   network technologies with connection-oriented behavior.


[Dhruv]: Yes that would be good with Eve suggestion and removing ‘in the document’/’of this draft’.


* “Transport networks have a variety of mechanisms to:
-               Facilitate separation of data plane and control plane,
-               Allow for distributed or centralized signaling  for path setup and protection, and
-               Provide traffic engineering mechanism via centralized path computation.”

Term ‘centralized signaling’ is confusing to me, do you mean, the use of NMS?

YOUNG>> Yes. We can reword the second point: “Allow for distributed signaling or an NMS-based centralized signaling for set up….”

[Dhruv]: That works!
----------------------------

* “The architecture work will lead to requirements for information models and protocol extensions between the virtual network controller and the physical network domain controllers and between the virtual network controller and multi-tenant customer controllers.”

It would be nice to add “multi-domain coordinator” with “virtual network controller” to explicitly link them together. Also multi-tenant customer controllers doesn’t convey the intention to have different customer controllers for each tenant, how about…

“The architecture work will lead to requirements for information models and protocol extensions between the virtual network controller (embedded in a multi-domain coordinator) and the physical network domain controllers and between the virtual network controller and multiple tenant customer controllers.”

YOUNG>> This sounds good to me. Thanks.

* “The working group will determine if new protocol extensions are necessary. If the working group determines they are necessary, then it will develop the new protocols within the working group where necessary, while interacting with other working groups to enhance existing protocols where possible.”

Suggest to re-word this as it’s not clear – is it about extension to existing protocols or new protocol and where that work might be taken up. Also Protocol extensions is mentioned as a work item after re-chartering.

YOUNG>> New protocol extensions mean a brand new protocol (say, “ACTN” protocol per se) that cannot be extended from existing protocols. But there may be some areas where we need to expand from existing protocols. In such case, we need to interact with the corresponding WGs. Perhaps an analogy would be that CCAMP WG is chartered to work on OPSF-TE for specific technologies while OSPF WG needs to be informed of the protocol changes.

[Dhruv]: How about we reword this to – “The working group will determine if new protocol or extensions to existing protocols are necessary. If the working group determines they are necessary, then it will develop new protocols within the working group, on the other hand any extensions to existing protocols would be done  with interactions with other working groups where possible.”

* You may want to do charter text formatting in RFC text format taking care of word-wraps and indentation
YOUNG>> Not sure what you meant here. Please clarify it for me.
[Dhruv]: Refer attachment, this format would help when we upload it to the data tracker and charter diff tool can help track it better.
This version has all the changes suggested by me and Eve.
Regards,
Dhruv
Hope you find them useful in making the charter text crisp.

YOUNG>> Definitely. Thanks a lot for your review and great comments.

Dhruv


From: ACTN [mailto:[e-mail]] On Behalf Of Leeyoung
Sent: 10 September 2014 04:53
To: Leeyoung; [e-mail]<mailto:[e-mail]>
Subject: [Actn] ACTN charter v1.0

Sorry, I forgot the Subject of this email. Here’s a retransmission with the subject.

From: Leeyoung
Sent: Tuesday, September 09, 2014 3:47 PM
To: [e-mail]<mailto:[e-mail]>
Subject:

Hi,

I hope your summer break was a good one.

We’d like to give you some updates and plans on the ACTN work. We are going to request a formal BoF in Honolulu IETF meeting. In doing so, we need a charter draft as part of the due diligence. Here’s an initial charter draft developed by the small set of proponents of the work based on the discussions and use-cases and other published documents.  We hope this captures a workable ACTN scope.  This version 1.0 draft charter is also posted in the wiki: https://sites.google.com/site/actnbof/home/charter-propor

Your review and comment will be greatly appreciated to come up with a good charter developed by the community of interest.

Best regards,
Young (on behalf of the proponents)

-------------------------------draft charter 1.0 ----------------------------------------------------------------------------------------------------

Transport networks have a variety of mechanisms to:

-       Facilitate separation of data plane and control plane,

-       Allow for distributed or centralized signaling  for path setup and protection, and

-       Provide traffic engineering mechanism via centralized path computation.

These represent key technologies for enabling flexible and dynamic networking, and efficient control and recovery of resources. Although these technologies provide significant benefits within a single domain control boundary, they do not meet the growing need for transport network virtualization in multi-domain transport networks. More and more network operators are building and operating on multi-domain transport networks. These domains (collections of links and nodes) may be each of a different technology, administrative zones, or vendor-specific islands. Establishment of end-to-end connections spanning multiple domains is a perpetual problem for operators because of both operational concerns (control plane and management plane) and interoperability issues (control plane and data plane).  Due to these issues, new services that require connections that traverse multiple domains need significant planning and often manual operations to interface different vendor equipment and technology.

The aim of Abstraction and Control of Transport Networks (ACTN) is to facilitate a centralized virtual network operation: the creation of a virtualized environment allowing operators to view and control multi-subnet, multi-technology, multi-vendor domain networks. Abstraction of transport networks also allows operators to consolidate their network services into multi-tenant virtual transport networks. This will enable rapid service deployment of new dynamic and elastic services, and will improve overall network operations and scaling of existing services. Discussion with operators has also highlighted a need for virtual network operation based on the abstraction of underlying technology and vendor domains.

Multi-domain network coordination function in ACTN is built on a control hierarchy where a multi-domain coordinator interacts with each domain controller (e.g., EMS/NMS, GMPLS/PCE control plane, SDN controller) for abstracting network resource information to provide virtual network control functions. This virtual network control functions are embedded in a multi-domain network coordinator to support various services/clients/applications to create and manage their own virtual networks that share the common transport network resources.


The ACTN working group will work to develop a high-level architecture for transport network abstraction and control that facilitates seamless vertical service coordination across multi-tenant customers (primarily internal service organizations with respect to a network operator), the virtual network control and the physical network domain controls as well as a horizontal E2E service coordination across multi-domain networks. It will identify key building components and the corresponding interfaces among these components.  The architecture work will lead to requirements for information models and protocol extensions between the virtual network controller and the physical network domain controllers and between the virtual network controller and multi-tenant customer controllers. Well-defined use cases from operators perspective with clearly stated need for transport network virtualization are critical in scoping the work and thus to achieve the deliverables of the working group.

The working group will work on the following items:

-       High-level architecture that describes the basic building blocks to enable transport network virtualization to support use cases.

-       Operator-driven use cases to address the following initial items:

o    Virtual network control and operation for core transport Packet Optical Integration (POI). (e.g., MPLS-TP, OTN/WSON)

o    Virtual network control and operation for mobile backhaul multi-technology transport (e.g., MPLS-TP and MPLS/OTN)

o    Data Center Operator’s interconnection with optical transport network infrastructure providers to support dynamic virtual circuit services.

o    Multi-tenant support to allow virtual network information query, virtual network negotiation, creation/deletion and modification.

o    Synchronization of network resources view across physical domain controls and virtual network control.

o    Dynamic service control and monitoring across all entities.

Initial work within the working group will be limited to a single operator administrative domain with an exception for the Data Center operation use case.

-       Evaluation of Information model/data model to support the use cases.

-       Requirements to support APIs/protocols, encoding languages, and data models

-       Gap analysis of existing IETF and other protocols, encoding languages and data models to fulfill the requirements.

-       Protocol extensions (if necessary after re-chartering).

The working group will determine if new protocol extensions are necessary. If the working group determines they are necessary, then it will develop the new protocols within the working group where necessary, while interacting with other working groups to enhance existing protocols where possible.




_______________________________________________
ACTN mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/actn



--
--
 =====================================
| Y. Richard Yang <[e-mail]<mailto:[e-mail]>>   |
| Professor of Computer Science       |
| http://www.cs.yale.edu/~yry/        |
 =====================================