ETSI GR NGP 006 V1.1.1 (2018-06)

Next Generation Protocols (NGP); Intelligence-Defined Network (IDN)

DGR/NGP-006

General Information

• Status: Published
• Publication Date: 14-Jun-2018
• Technical Committee: NGP - Next Generation Protocols
• Current Stage: 12 - Completion
• Due Date: 03-Jul-2018
• Completion Date: 15-Jun-2018

Standards Content (Sample)

ETSI GR NGP 006 V1.1.1 (2018-06)






GROUP REPORT
Next Generation Protocols (NGP);
Intelligence-Defined Network (IDN)
Disclaimer
The present document has been produced and approved by the Next Generation Protocols (NGP) ETSI Industry Specification
Group (ISG) and represents the views of those members who participated in this ISG.
It does not necessarily represent the views of the entire ETSI membership.

---------------------- Page: 1 ----------------------
2 ETSI GR NGP 006 V1.1.1 (2018-06)



Reference
DGR/NGP-006
Keywords
framework, next generation protocol

ETSI
650 Route des Lucioles
F-06921 Sophia Antipolis Cedex - FRANCE

Tel.: [phone]  Fax: [phone]

Siret N° 348 623 562 00017 - NAF 742 C
Association à but non lucratif enregistrée à la
Sous-Préfecture de Grasse (06) N° 7803/88

Important notice
The present document can be downloaded from:
http://www.etsi.org/standards-search
The present document may be made available in electronic versions and/or in print. The content of any electronic and/or
print versions of the present document shall not be modified without the prior written authorization of ETSI. In case of any
existing or perceived difference in contents between such versions and/or in print, the only prevailing document is the
print of the Portable Document Format (PDF) version kept on a specific network drive within ETSI Secretariat.
Users of the present document should be aware that the document may be subject to revision or change of status.
Information on the current status of this and other ETSI documents is available at
https://portal.etsi.org/TB/ETSIDeliverableStatus.aspx
If you find errors in the present document, please send your comment to one of the following services:
https://portal.etsi.org/People/CommiteeSupportStaff.aspx
Copyright Notification
No part may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying
and microfilm except as authorized by written permission of ETSI.
The content of the PDF version shall not be modified without the written authorization of ETSI.
The copyright and the foregoing restriction extend to reproduction in all media.

© ETSI 2018.
All rights reserved.

TM TM TM
DECT , PLUGTESTS , UMTS and the ETSI logo are trademarks of ETSI registered for the benefit of its Members.
TM TM
3GPP and LTE are trademarks of ETSI registered for the benefit of its Members and
of the 3GPP Organizational Partners.
oneM2M logo is protected for the benefit of its Members.
®
GSM and the GSM logo are trademarks registered and owned by the GSM Association.
ETSI

---------------------- Page: 2 ----------------------
3 ETSI GR NGP 006 V1.1.1 (2018-06)
Contents
Intellectual Property Rights . 4
Foreword . 4
Modal verbs terminology . 4
1 Scope . 5
2 References . 5
2.1 Normative references . 5
2.2 Informative references . 5
3 Abbreviations . 6
4 Overview . 6
5 Background . 6
5.1 Continuous Evolution of Network . 6
5.2 Functional and Systemic Requirement . 7
5.3 Rapid Development of Machine Learning Technologies . 8
6 Benefits of Introducing AI into Network . 9
6.1 Towards Fully Autonomic Network . 9
6.2 Response to the challenge of complexity . 9
6.3 Response to the challenge of variation . 10
6.4 Insights of the Network and Improve the Utilization . 10
6.5 To Be Predictive . 11
6.6 Potential Decision Efficiency . 12
6.7 Potential Business Model . 12
7 Design Goals of IDN . 12
7.1 Goal of IDN . 12
7.2 Deployment models: Centralized, distributed and Hybrid . 13
7.3 Wired and wireless consideration . 14
7.4 Security and Privacy Considerations . 16
7.5 Multi-objectives Resolution . 17
8 The proposed IDN Architecture . 17
8.1 Reference Architecture . 17
8.2 Comparing System design . 21
8.2.1 Overview . 21
8.2.2 Distributed Architecture . 23
8.2.3 Centralized Architecture . 23
8.2.4 Hybrid Architecture . 24
8.3 Controlling Loop . 25
8.3.1 AI-Enhanced Close Loop . 25
8.3.2 AI-Enhanced Open Loop . 27
8.3.3 Traditional Loop . 29
8.3.4 Internal Loop . 29
8.3.5 UNI Loop . 29
8.4 Core Support Technologies . 29
8.4.1 Network modelling . 29
8.4.2 Measurement and Data Orchestration . 30
9 Potential Standardization Works . 31
9.1 Overview . 31
9.2 Measurement . 32
9.3 Data Centric standards . 32
9.4 Control Centric standards . 33
Annex A: Authors & contributors . 34
History . 35

ETSI

---------------------- Page: 3 ----------------------
4 ETSI GR NGP 006 V1.1.1 (2018-06)
Intellectual Property Rights
Essential patents
IPRs essential or potentially essential to normative deliverables may have been declared to ETSI. The information
pertaining to these essential IPRs, if any, is publicly available for ETSI members and non-members, and can be found
in ETSI SR 000 314: "Intellectual Property Rights (IPRs); Essential, or potentially Essential, IPRs notified to ETSI in
respect of ETSI standards", which is available from the ETSI Secretariat. Latest updates are available on the ETSI Web
server (https://ipr.etsi.org/).
Pursuant to the ETSI IPR Policy, no investigation, including IPR searches, has been carried out by ETSI. No guarantee
can be given as to the existence of other IPRs not referenced in ETSI SR 000 314 (or the updates on the ETSI Web
server) which are, or may be, or may become, essential to the present document.
Trademarks
The present document may include trademarks and/or tradenames which are asserted and/or registered by their owners.
ETSI claims no ownership of these except for any which are indicated as being the property of ETSI, and conveys no
right to use or reproduce any trademark and/or tradename. Mention of those trademarks in the present document does
not constitute an endorsement by ETSI of products, services or organizations associated with those trademarks.
Foreword
This Group Report (GR) has been produced by ETSI Industry Specification Group (ISG) Next Generation
Protocols (NGP).
Modal verbs terminology
In the present document "should", "should not", "may", "need not", "will", "will not", "can" and "cannot" are to be
interpreted as described in clause 3.2 of the ETSI Drafting Rules (Verbal forms for the expression of provisions).
"must" and "must not" are NOT allowed in ETSI deliverables except when used in direct citation.

ETSI

---------------------- Page: 4 ----------------------
5 ETSI GR NGP 006 V1.1.1 (2018-06)
1 Scope
The scope of the present document is to specify the self-organizing control and management planes for the Next
Generation Protocols (NGP), Industry Specific Group (ISG).
2 References
2.1 Normative references
Normative references are not applicable in the present document.
2.2 Informative references
References are either specific (identified by date of publication and/or edition number or version number) or
non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the
referenced document (including any amendments) applies.
NOTE: While any hyperlinks included in this clause were valid at the time of publication, ETSI cannot guarantee
their long term validity.
The following referenced documents are not necessary for the application of the present document but they assist the
user with regard to a particular subject area.
[i.1] https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-
it-racism/.
[i.2] https://www.thesun.co.uk/tech/4141624/facebook-robots-speak-in-their-own-language/.
[i.3] Reed S, Akata Z, Yan X, et al.: "Generative adversarial text to image synthesis", in ICML 2016.
[i.4] Oord A, Dieleman S, Zen H, et al.: "Wavenet: A generative model for raw audio",
arXiv:1609.03499, 2016.
[i.5] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton: "Deep learning", in Nature 521.7553 (2015):
436-444.
[i.6] Kingma D P, Welling M.: "Auto-encoding variational bayes", in ICLR 2014.
[i.7] Goodfellow, Ian, et al.: "Generative adversarial nets", in NIPS 2014.
[i.8] Cisco White Paper.
NOTE: Available at https://www.cisco.com/c/en/us/products/collateral/routers/wan-automation-
engine/white_paper_c11-728552.html.
[i.9] https://arxiv.org/abs/1701.07274.
[i.10] ETSI TR 121 905: "Digital cellular telecommunications system (Phase 2+) (GSM); Universal
Mobile Telecommunications System (UMTS); LTE; Vocabulary for 3GPP Specifications
(3GPP TR 21.905)".
[i.11] ETSI TS 136 401: "LTE; Evolved Universal Terrestrial Radio Access Network (E-UTRAN);
Architecture description (3GPP TS 36.401)".
ETSI

---------------------- Page: 5 ----------------------
6 ETSI GR NGP 006 V1.1.1 (2018-06)
3 Abbreviations
For the purposes of the present document, the abbreviations given in ETSI TR 121 905 [i.10] and the following apply to
scenarios that include mobile network architectures:
TM rd
3GPP 3 Generation Participation Project
AI Artificial Intelligence
DHCP Dynamic Host Configuration Protocol
E-W East and West (direction)
IDN Intelligence-Defined Network
IETF Internet Engineering Task Force
IP Internet Protocol
ISG Industry Specific Group
ML Machine Learning
NE Network Element
NGP Next Generation Protocols
NMS Network Management System
N-S North and South (direction)
OAM Operation And Management
OSPF Open Shortest Path First
QoE Quality of Experience
4 Overview
The Next Generation Protocols (NGP), ISG aims to review the future landscape of Internet Protocols, identify and
document future requirements and trigger follow up activities to drive a vision of a considerably more efficient Internet
that is far more attentive to user demand and more responsive whether towards humans, machines or things.
A measure of the success of NGP would be to remove historic sub-optimised IP protocol stacks and allow all next
generation networks to inter-work in a way that accelerates a post-2020 connected world unencumbered by past
developments.
The NGP ISG is foreseen as having a transitional nature that is a vehicle for the 5G community and other related
communications markets to first gather their thoughts together and prepare the case for the Internet community's
engagement in a complementary and synchronised modernisation effort.
Therefore NGP ISG aims to stimulate closer cooperation over standardisation efforts for generational changes in
communications and networking technology.
The present document focuses on proposing a new Intelligence-Defined Network (IDN) architecture and a gap analysis
of current architectures. The intelligence technologies can learn from historical data, and make predictions or decisions,
rather than following strictly predetermined rules. On one hand, the IDN can dynamically adapt to a changing situation
and enhance its own intelligence with by learning from new data. On the other hand, IDN can also aim at supporting
human-based decision by pre-processing data and rendering insights to users through advanced user interfaces and
visualisations. The integration with various network infrastructures, such as SDN, NFV&MANO, intelligence router,
traditional router, is in the scope of the present document.
5 Background
5.1 Continuous Evolution of Network
The development of network is continuously evolving process. In different stages, the network faces to various and
different complexity problems. Therefore, the operating and management methodologies are also various.
ETSI

---------------------- Page: 6 ----------------------
7 ETSI GR NGP 006 V1.1.1 (2018-06)

Figure 1: Three Stages Development
As Figure 1 shows, early networks are referred as Manually Defined Network. In such networks, the OAM basic
approach is network planning, CLI configuration and network optimization. All the operations are fully human driven.
Since the administrator needed to configure and control each device individually, the complexity and the cost of OAM
was very high.
Along with the development, the scale of network and service became larger and larger. The OAM requirement has also
increased. Due to the high degree of required control degree of requirement, a virtual control layer was developed. This
layer supports batch operation of low layer devices, which improves the efficiency significantly. Because of the divide
of control layer and forwarding layer, the configuration and controlling operation is implemented by south and north-
bound cooperation. Southbound typically uses Netconf/YANG, OpenFlow, etc. to configure the network forwarding,
policies, etc. Northbound abstracts the functionalities for application requirements, thus deriving the forwarding table
and policies, etc. With this paradigm, the entire network was transformed to become semi-automatic. Many operations
can be executed automatically and the administrator is only responsible for decision making.
Currently, the network undergoing a new transformation towards Intelligence-Defined Network (IDN). Since the
network problems evolve to more complex, the traditional human decision-making can hardly support the requirements.
Therefore, the AI methods, which can help for decision making and analysis, are introduced to solve OAM problems.
The core of IDN is machine learning algorithms and models. The network, traffic and application patterns can be
modelled by AI methods via learning from the existing data and experiences. It is expected as a full-autonomic system
that can make decision itself, especially in the common and repeat events that do not need human to judge. This will
decrease the OAM cost hugely in the future.
5.2 Functional and Systemic Requirement
IDN is seen as the next form of network evolution. Comparing with the current state, there are new requirements that
declare the essential improvements of new approach.
The first part is functional requirements, which means the IDN approach should own the functions that the previous
network approaches do not have. For IDN, some of the potential functional requirements are following:
• Real-time assessment. The IDN approach should provide a consolidated view of the current network status
including traffic and running applications by providing aggregated and condensed insights.
• Prediction or inference. The IDN approach should have the ability to predict / infer the oncoming trend of
network in multiple dimensions, such as inferring the QoS parameter according to the traffic matrix. This
ability will support the intelligent system to implement proactive operations.
• Autonomic decision making. The network will not only execute the policies which produced by administrator
but also autonomously make decision. This is one of the most important reasons that why AI technology is
introduced into network.
ETSI

---------------------- Page: 7 ----------------------
8 ETSI GR NGP 006 V1.1.1 (2018-06)
• Verification. IDN is not only in charge of taking decisions but also (1) verifies that its own decision are
properly applied and results in the expected states and (2) can be leveraged to verify that policies derived from
multiple entities (concurrent IDN algorithms and even users) are coherent.
• Dynamical configuring ability. For stability consideration, typically operators try to minimize changes on the
devices. However, one of the purposes of introducing AI technology is to modify the configuration so that
adapt for the variation of network traffic and state.
The second part is systemic requirements, which means the IDN approach should own the system level abilities in the
low layer (or say primordially) that the previous network approaches do not have or cannot easily complement. For
IDN, the potential systemic requirements are following but not limited.
• Inherent data collection and orchestration. The current measure method is driven by external command.
Namely, all the network data is a response (or feedback) of a specific command. The network devices do not
widely support the actively data upload functions. This leads to at least two problems. The first one is the cost.
When the measured data volume is large, there is nearly half of the signal messages and transfer time are
wasted because one data feedback should be potentially triggered by one measure signal. This external trigger
mechanism may not satisfy the requirements of huge network data collection. The second one is the
complexity. The current measured factors are few (delay, jitter, loss). Even if in this case, it is hard to obtain
the accurate data according to simple operations. The intelligent system may handle not only the existing
factors but also other complex data types. Some of the factors may be hard to measure, such as if the queuing
length is wanted to know. Furthermore, IDN decision algorithms could also rely on external data for which
particular connectors are required. This potentially becomes one of the key systemic requirements.
• Data pre-processing. As multiple sources of data will be leveraged, normalisation techniques it its large sense
should be used (including data alignment, sanitization). It also concerns the establishment of proper metrics
(distance, similarities, and dissimilarities) which are in the core of ML algorithms whereas some collected data
may not be easily mapper to a metric space by nature.
• Map algorithm to network. There is a huge gap between the current AI algorithms input/output and network
policy. The former is pure mathematical expression while the latter one tends to be a kind of programmed
language. If the intelligent system is seen as the mapping of physical network, it is very important to build up
the "bridge" between the network semantics and algorithm semantics. Different with the process of data
orchestration, the core problems here is how to generate and delivery the network policies based on the
mathematical input/output of the algorithm.
5.3 Rapid Development of Machine Learning Technologies
Even though the use of Machine Learning technologies is still in its infancy in most fields of networking, it will become
a much thought for opportunity to enhance network operations and performance in the coming years. This is mostly due
to the rapid development of Machine Learning (ML) and associated Artificial Intelligence (AI) technologies in other
fields.
ML/AI in picture/video/speech recognition as well as big-data analytics in areas such as e-commerce and search have
evolved to a point where many of the methods and components of building solutions are well enough understood to
apply them to novel fields - such as networking.
The ability of developers to rapidly build systems with ML/AI was vastly improved in the last few years through
common tools such as TensorFlow that took most of the novel and unique complexity of building ML/AI solution into
those expert built tools/libraries. The layers above those common libraries now become areas of development where
more and more the subject matter experts (such as networking engineers) will be able to collaborate with data analysts
to build those ML/AI solutions.
The performance of both AI/ML learning/training as well as the execution of the trained neural networks has been
improved radically in the past years and it is expected lot more of these recent developments to proliferate into
products.
GPUs (Graphic Processor Units) such as those from NVidia (as leader in the market) have evolved to be equal good
high-performance parallel execution units for ML/AI training and inference. Algorithms to improve performance of
execution by more than a factor of 1 000 have been developed in the past years.
ETSI

---------------------- Page: 8 ----------------------
9 ETSI GR NGP 006 V1.1.1 (2018-06)
Low-end ML/AI neural network inference hardware is now being released on products. Product means that these are
hardware building block that can for example be added to existing low-end CPU chips such as ARM CPUs for
cellphones and low-end network devices. This hardware can only execute neural networks (this is called inference), but
not train those neural networks. These accelerators can do inference at minute fractions of the power needed in GPUs.
The likely first big area where these will be used is speech recognition and translation on mobile phones.
6 Benefits of Introducing AI into Network
6.1 Towards Fully Autonomic Network
A fully autonomic network means that the network contains a closed-loop of "Measure-Analyse-Decide" which can
realize the whole process autonomically. By means of AI-based learning and optimization techniques, the goal of IDN
architecture is to learn about its behaviour, the fundamental relation between traffic load, network configuration and the
resulting performance, understand the target policy set by the network administration, and configure that policy
efficiently and fully autonomously. The advantage of a fully autonomic network is realizing the closed-loop of
"Measure-Analyse-Decide", which will minimize the requirements for human administrators.
Currently, the process of measure, analysis and decision are mainly independent and the cooperation of such processes
typically relies on humans. The limitation is caused by the lack of analysis ability of network, which is precisely AI
technology performs really well. While in operation, the IDN architecture will react autonomously to relevant events
(e.g. a failure, a spike in the traffic, etc.) and change the configuration accordingly. The core of AI technology is
extracting the patterns (or knowledge) from complex data, in other words, discovering the rules then applying. In
current, the forwarding process has achieved stateless or stateful full automation while most of the controlling process,
such as the configuration and optimization, are still manual. The roadmap should be gradual, which starts from the local
area autonomic to large area and finally to global. As if the development of self-driving, the automatic transformation is
realized step by step, from such as auto-shift and auto-break. As yet, the AI technology is the one of the most possible
ways to realize the whole process. During the development, the introducing of the AI technology gradually implement
the closed-loop of network controlling so that reduce the unnecessary manual operation including coding, configuring,
simple inference, etc. A fully autonomic network potentially decreases the cost of carriers during management and
control. It will be benefit for the income in the long term.
6.2 Response to the challenge of complexity
AI and most notably Machine Learning (ML) techniques play a central role in the future architecture of networks. By
means of ML mechanisms, the network behaviour can be learnt to obtain a ML-based model. This model can account
for any arbitrary network characteristic of interest. As examples the models can characterize the energy consumption of
the network or understand the relation between the traffic load and external factors such as popular sports events.
Traditionally network modelling has been done by means of simulation, however ML provides many advantages in this
regard. First, ML scales very well with complexity and it is able to understand and model non-linear (complex) issues,
indeed deep neural networks are able to account for multi-dimensional non-linear problems. On the contrary,
simulations require costly development to model complex behaviour. Second, although training the neural networks is a
CPU/GPU intensive process, once trained the neural network is ve
...