Re: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

["Dongjie (Jimmy)" <[e-mail]>]

Hi Krzysztof,

Yes NRP is a concept introduced for the realization of IETF network slice, and it is the result of long time WG discussion about the terminology for the underlay network construct which is used to support the IETF network slice services. My personal understanding is that the NRP concept and definition belongs to the framework document, while the technologies used to instantiate NRP are out of its scope. For sure there can be multiple ways to build an NRP, as long as they comply to the NRP definition in the framework.

I understand your concern is whether the framework mandates only one or two mechanisms to build NRP, I don’t think this is what the current framework indicates. Thus to me the examples about detailed realization mechanisms are not quite necessary for the framework document and may contradict with its role as a high-level framework.

On the other hand, discussion about the possible mechanisms to build NRPs can continue in the WG.

To me an NRP is usually realized with the combination of a set of data plane and control plane technologies. Constrained path computation or Flex-Algo can only be seen as a component rather than a complete solution. As you mentioned, one or multiple technologies such as admission control/advanced or basic QoS/capacity planning/resource reservation either on the edge nodes or the transit nodes would be needed, and the mechanisms can have specific applicability to different network applications.

Best regards,
Jie

From: Krzysztof Szarkowicz [mailto:[e-mail]]
Sent: Thursday, April 14, 2022 9:00 PM
To: Dongjie (Jimmy) <[e-mail]>
Cc: John E Drake <[e-mail]>; [e-mail]; <[e-mail]> <[e-mail]>; [e-mail]
Subject: Re: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

Hi Jie,

I agree, NRP is a realization, and a such, my initial comments (few weeks ago) was that it should not be mentioned at all in the framework document, but should be left to the documents describing the realization.

If, however, NRP is mandated by the framework document for realization, the wording for NRP in the framework document should be wide enough, to allow different realization options (described in different realization documents). For example, realization option mentioned by Med:

* admission control with advanced QoS (large number of edge QoS classes) at the transport edge
* basic QoS (limited number of transport core QoS classes) in transport core
* mapping between large number of transport edge QoS classes to limited number transport core QoS classes
* capacity planning

has no good match with current wording about NRP in the framework document, IMHO.

Please see as well inline.

Thanks,
Krzysztof



On 2022 -Apr-14, at 09:05, Dongjie (Jimmy) <[e-mail]<mailto:[e-mail]>> wrote:

Hi John,

Thanks for the discussion on this point. Please see further inline with [Jie]:

From: John E Drake [mailto:[e-mail]]
Sent: Wednesday, April 13, 2022 9:44 PM
To: Dongjie (Jimmy) <[e-mail]<mailto:[e-mail]>>; Krzysztof Szarkowicz <[e-mail]<mailto:[e-mail]>>; [e-mail]<mailto:[e-mail]>; <[e-mail]<mailto:[e-mail]>> <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>
Subject: RE: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

Hi,

Upon re-reading Krzysztof’s email, I think we should keep the existing text and supplement it with Krzysztof’s text, but striking the second sentence.  Comments inline below.

Yours Irrespectively,

John


Juniper Business Use Only
From: Dongjie (Jimmy) <[e-mail]<mailto:[e-mail]>>
Sent: Wednesday, April 13, 2022 2:53 AM
To: Krzysztof Szarkowicz <[e-mail]<mailto:[e-mail]>>; [e-mail]<mailto:[e-mail]>; John E Drake <[e-mail]<mailto:[e-mail]>>; <[e-mail]<mailto:[e-mail]>> <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>
Subject: RE: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

[External Email. Be cautious of content]

Hi Krzysztof,

One concern is the text you proposed is the level of detail which the framework document was trying to avoid.

And as for the examples you provided, I’m not sure if all of them can be called NRP. For example:

a)       collection of paths optimized based on certain criteria. For example low latency paths for NSP-A, and high capacity paths for NSP-B.

Low latency paths and high capacity paths may be obtained by using different metric types in Flex-Algo or other path computation, while as mentioned in a previous mail, depends on the network structure and topology, the path computed based on different metric types or constraints may still be the same or contain a set of shared links.

[JD]  What you are describing is how the NRP specified in a), above, might be instantiated.  I think a) is consistent with the existing text

[Jie] My point was that the mechanism in a) may not be a good example for NRP instantiation. As it is only path computation and does not describe how the resources are managed or reserved for different NRPs.

Then it is not clear how the performance of services mapped to different NRPs with shared paths or links can be guaranteed. IMO some kind of resource reservation or management is needed for different NRPs.

[JD]  This is an issue for the service provider

[Jie] Indeed service provider needs the mechanisms which can create NRPs to meet the performance requirement of different services, even if the paths or links are shared with other services.

[Krzysztof] This is the question of realization, not the question of the framework. Some realization might use some advanced resource reservation/management for each link in the network, to squeeze out the link to last possible bit, some other realization might just use advanced reservation techniques on some links only (for example transport edge links), while using more loosely techniques on other links (e.g. transport core links). Some other realization might over provision certain links, instead of using advanced reservation mechanisms on these links. And, there could be dozens of different realization options, optimized for specific underlying technology. The framework document should not dictate any realization option, IMHO. It should be left to the realization documents.



Best regards,
Jie


From: Teas [mailto:[e-mail]] On Behalf Of Krzysztof Szarkowicz
Sent: Tuesday, April 12, 2022 10:08 PM
To: [e-mail]<mailto:[e-mail]>; John E Drake <[e-mail]<mailto:[e-mail]>>; <[e-mail]<mailto:[e-mail]>> <[e-mail]<mailto:[e-mail]>>
Cc: [e-mail]<mailto:[e-mail]>
Subject: Re: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

John, Adrian, Med

In that case, I would propose following text update. The reason is, from the current text, I cannot extract the the examples provided by Med and Adrian could be and well considered as an NSP realization.

===== Current Text ============
An NRP is simply a collection of resources identified in the underlay network.  Thus, the NRP is a scoped view of a topology and may be considered as a topology in its own right.  The process of determining the NRP may be made easier if the underlay network topology is first filtered into a Filter Topology in order to be aware of the subset of network resources that are suitable for specific NRPs, but this is optional.
====== End of current text ==========


===== New Text ============
It is out of scope for this document to describe the details, how an NSP can be realized. However, at a high-level, an NRP is simply a collection of resources identified in the underlay network. Some (none exhaustive) examples of an NSP realization are:

a) collection of paths optimized based on certain criteria. For example low latency paths for NSP-A, and high capacity paths for NSP-B.

b) admission control - possibly with large number of QoS classes - at the transport access, coupled with limited set of transport core QoS classes (PDBs), and N:M mapping between access and core classes.

c) combination of above

d) more complex realization schemes, with combinations of advanced QoS at both the access and transport core, possibly coupled with Filter Topology and different path optimizations methods.
====== End of new text ==========

Best regards,
Krzysztof Szarkowicz

On 2022 -Apr-01, at 14:43, John E Drake <[e-mail]<mailto:[e-mail]>> wrote:

Krzysztof,

The usual answer to such a request is to ask you to provide text.

Yours Irrespectively,

John





On 2022 -Apr-01, at 15:44, [e-mail]<mailto:[e-mail]> wrote:

Re-,

Indeed.

FWIW, some of this is captured in RFC7297 with a focus on legacy mechanisms:

==
   These requirements include: reachability scope (e.g., limited scope,
   Internet-wide), direction, bandwidth requirements, QoS parameters
   (e.g., one-way delay [RFC2679], loss [RFC2680], or one-way delay
   variation [RFC3393]), protection, and high-availability guidelines
   (e.g., restoration in less than 50 ms, 100 ms, or 1 second).

   These requirements are then translated into IP/MPLS-related technical
   clauses (e.g., need for recovery means, definition of the class of
   service, need for control-plane protection, etc.).  In a later stage,
   these various clauses will be addressed by the activation of adequate
   network features and technology-specific actions (e.g., Multiprotocol
   Label Switching Traffic Engineering (MPLS-TE, [RFC3346]), Resource
   Reservation Protocol (RSVP, [RFC2205]), Open Shortest Path First
   (OSPF), Intermediate System to Intermediate System (IS-IS), etc.), by
   means of CPP-derived configuration information.

   …

   The CPP template aims to capture connectivity needs and to represent
   and value these requirements in a standardized manner.  Service- and
   Customer-specific IP provisioning rules may lead to a dramatic
   increase of the number of IP transfer classes that need to be
   (pre-)engineered in the network.  Instantiating each CPP into a
   distinct class of service should therefore be avoided for the sake of
   performance and scalability.

   Therefore, application-agnostic IP provisioning practices should be
   recommended, since the requirements captured in the CPP can be used
   to identify which network class of service is to be used to meet
   those requirements/guarantees.  From that standpoint, the CPP concept
   is meant to design a limited number of generic classes so that
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   individual CPP documents, by capturing the connectivity requirements
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   of services, applications, and Customers, can be easily mapped to
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   these classes.
   ^^^^^^^^^^^^
==

Cheers,
Med

De : Krzysztof Szarkowicz <[e-mail]<mailto:[e-mail]>>
Envoyé : vendredi 1 avril 2022 15:26
À : BOUCADAIR Mohamed INNOV/NET <[e-mail]<mailto:[e-mail]>>
Cc : [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Objet : Re: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

Med,

We are on the same page here. I see this as well as predominant approach, in fact.

Cheers,
Krzysztof


On 2022 -Apr-01, at 15:20, <[e-mail]<mailto:[e-mail]>> <[e-mail]<mailto:[e-mail]>> wrote:

Krzysztof,

Admission control + use of limited set of core QoS classes (PDBs) + much more access QoS classes + map access QoS classes to core classes + no multi-topology/customized path selection is also an example network partitioning. I even expect this to be the privileged approach when legacy forwarding mechanisms should be used to fulfil slicing objectives.

Cheers,
Med

De : Krzysztof Szarkowicz <[e-mail]<mailto:[e-mail]>>
Envoyé : vendredi 1 avril 2022 10:30
À : BOUCADAIR Mohamed INNOV/NET <[e-mail]<mailto:[e-mail]>>; [e-mail]<mailto:[e-mail]>
Cc : [e-mail]<mailto:[e-mail]>
Objet : Re: [Teas] [E] Re: Slicing Framework Open issue #1 : Service != Realization

Adrian, Med,

Returning to this. I’ve seen many design cases with SPs, where the network is designed physically in very structured manner. Also, I’ve seen some analysis done by couple of SPs, that as the result of this very structured network design, after making internal analysis, they don’t really see the difference between path placement done using IGP metric, and path placement done using delay metric. Hence, even introducing some flex-algos (TE meshes) with different metric optimization is questionable here (as the resulting paths are the same).

To realize services with SLO commitments, these SPs today extensively use admission control at the edge (some times even relatively complex admission control schemes at the edge). But, they don’t use anything specific (TE, Flex-Algo) in the transport, apart from mentioned admission control at the edge (ingress/egress policers/shapers), basic QoS on transit, network capacity planning, and timely capacity upgrades, where needed.

Does this approach could be considered as well as some way of ’network resource partitioning’ mentioned in the draft?

Cheers,
Krzysztof


On 2022 -Mar-28, at 12:26, Krzysztof Szarkowicz <[e-mail]<mailto:[e-mail]>> wrote:

OK,

If a realization example of network partition is flex-algo + admission control, then I am fine.

Thanks,
Krzysztof



_________________________________________________________________________________________________________________________



Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc

pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler

a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,

Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.



This message and its attachments may contain confidential or privileged information that may be protected by law;

they should not be distributed, used or copied without authorisation.

If you have received this email in error, please notify the sender and delete this message and its attachments.

As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.

Thank you.