Network Working Group C. Li
Internet Draft China Telecom
Intended status: Informational Y. Cheng
Expires: January 2020 China Unicom
J. Strassner
O. Havel
W. Liu
Huawei Technologies
P. Martinez-Julia
NICT
J. Nobre
UFRGS
D. Lopez
Telefonica I+D
July 9, 2019
Intent Classification
draft-li-nmrg-intent-classification-01
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at
http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
at any time. It is inappropriate to use Internet-Drafts as
reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 8, 2009.
Liu, et al. Expires January 8, 2020 [Page 1]
Internet-Draft Intent Classification July 2019
Copyright Notice
Copyright (c) 2019 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with
respect to this document. Code Components extracted from this
document must include Simplified BSD License text as described in
Section 4.e of the Trust Legal Provisions and are provided without
warranty as described in the Simplified BSD License.
Abstract
RFC7575 defines Intent as an abstract high-level policy used to
operate the network. Intent management system includes an interface
for users to input requests and an engine to translate the intents
into the network configuration and manage their lifecycle. Up to
now, there is no commonly agreed definition, interface or model of
intent.
This document discusses what intent means to different stakeholders,
describes different ways to classify intent, and an associated
taxonomy of this classification. This is a foundation for discussion
intent related topics.
Table of Contents
1. Introduction ................................................ 3
2. Acronyms .................................................... 3
3. Abstract intent requirements ................................. 4
3.1. What is Intent? ......................................... 4
3.2. Intent Solutions & Intent Users ......................... 4
3.3. Current Problems & Requirements ......................... 5
3.4. Intent Types that need to be supported .................. 7
4. Functional Characteristics and Behavior ...................... 8
4.1. Persistence ............................................ 8
4.2. Granularity ............................................ 9
4.3. Hierarchy .............................................. 9
4.4. Abstracting Intent Operation ........................... 10
4.5. Policy Subjects and Policy Targets ..................... 11
4.6. Policy Scope .......................................... 11
5. The Policy Continuum ........................................ 12
Li, et al. Expires January 8, 2020 [Page 2]
Internet-Draft Intent Classification July 2019
6. Involvement of intent in the application of AI to Network Manage
ment .......................................................... 12
7. Security Considerations ..................................... 13
8. IANA Considerations ......................................... 13
9. Contributors ................................................ 13
10. Acknowledgments ............................................ 13
11. References ................................................. 14
11.1. Normative References ................................. 14
11.2. Informative References ................................ 14
1. Introduction
Different SDOs (such as [ANIMA][ONF][ONOS]) have proposed intent as
a declarative interface for defining a set of network operations to
execute.
Although there is no common definition or model of intent which are
agreed by all SDOs, there are several shared principles:
o intent should be declarative, using and depending on as few
deployment details as possible and focusing on what and not how
o intent should provide an easy-to-use interface, and use
terminology and concepts familiar to its target audience
o intent should be vendor-independent and portable across
platforms
o the intent framework should be able to detect and resolve
conflicts between multiple intents.
SDOs have different perspectives on what intent is, what set of
actors it is intended to serve, and how it should be used. This
document provides several dimensions to classify intents.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].
2. Acronyms
CLI: Command Line Interface
SDO: Standards Development Organization
SUPA: Simplified Use of Policy Abstractions
Li, et al. Expires January 8, 2020 [Page 3]
Internet-Draft Intent Classification July 2019
VPN: Virtual Private Network
DC: Data Center
3. Abstract intent requirements
In order to understand the different intent requirements that would
drive intent classification, we first need to understand what intent
means for different intent users.
3.1. What is Intent?
The term Intent has become very widely used in the industry for
different purposes, sometimes it is not even in agreement with SDO
shared principles mentioned in the Introduction. Different
stakeholders consider an intent to be an ECA policy, a GBP policy, a
business policy, a network service, a customer service, a network
configuration, application / application group policy, any
operator/administrator task, network troubleshooting / diagnostics /
test, a new app, a marketing term for existing
management/orchestration capabilities, etc. Their intent is
sometimes technical, non-technical, abstract or technology specific.
For some stakeholders, intent is a subset of these and for other
stakeholders intent is all of these. It has in some cases become a
term to replace a very generic 'service' or 'policy' terminology.
While it is easier for those familiar with different standards to
understand what service, CFS, RFS, resource, policy continuum, ECA
policy, declarative policy, abstract policy or intent policy is, it
may be more difficult for the wider audience. Intent is very often
just a synonym for policy. Those familiar with policies understand
the difference between a business, intent, declarative, imperative
and ECA policy. But maybe the wider audience does not understand the
difference and sometimes equates the policy to an ECA policy.
Therefore, it is important to start a discussion in the industry
about what intent is for different solutions and intent users. It is
also imperative to try to propose some intent categories /
classifications that could be understood by a wider audience. This
would help us define intent interfaces, DSLs and models.
3.2. Intent Solutions & Intent Users
Different Solutions and Actors have different requirements,
expectations and priorities for intent driven networking. They
require different intent types and have different use cases. Some
users are more technical and require intents that expose more
Li, et al. Expires January 8, 2020 [Page 4]
Internet-Draft Intent Classification July 2019
technical information. Other users do not understand networks and
require intents that shield them from different networking concepts
and technologies. The following are the solutions and intent users
that intent driven networking needs to support:
+--------------------+------------------------------------+
| Solutions | Intent Users |
+--------------------+------------------------------------+
| Carrier Networks | Network Operator |
| | Service Designers |
| | Service Operators |
| | Customers/Subscribers |
+--------------------+------------------------------------+
| DC Networks | Cloud Administrator |
| | Underlay Network Administrator |
| | App Developers |
| | End Users |
+--------------------+------------------------------------+
| Enterprise Networks| Enterprise Administrator |
| | App Developers |
| | Enterprise Administrator |
+--------------------+------------------------------------+
3.3. Current Problems & Requirements
Network APIs and CLIs are too complex due to the fact that they
expose technologies & topologies. App developers and end-users do
not want to set IP Addresses, VLANs, subnets, ports, etc. Operators
and administrators would also benefit from the simpler interfaces,
like:
o Allow Customer Site A to be connected to Internet via Network B
o Allow User A to access all internal resources, except the Server
B
o Allow User B to access Internet via Corporate Network A
o Move all Users from Corporate Network A to the Corporate Network
B
o Request Gold VPN service between my sites A, B and C
o Provide CE Redundancy for all Customer Sites
Li, et al. Expires January 8, 2020 [Page 5]
Internet-Draft Intent Classification July 2019
o Add Access Rules to my Service
Networks are complex, with many different protocols and
encapsulations. Some basic questions are not easy to answer:
o Can User A talk to User B?
o Can Host A talk to Host B?
o Are there any loops in my network?
o Are Network A and Network B connected?
o Can User A listen to communications between Users B & C?
Operators and Administrators manually troubleshoot and fix their
networks and services. They instead want:
o a reliable network that is self-configured and self-assured based
on the intent
o to be notified about the problem before the user is aware
o automation of network/service recovery based on intent (self-
healing, self-optimization)
o to get suggestions about correction/optimization steps based on
experience (historical data & behaviour)
Therefore, Operators and Administrators want to:
o simplify and automate network operations
o simplify definitions of network services
o provide simple customer APIs for Value Added Services (operators)
o be informed if the network or service is not behaving as
requested
o enable automatic optimization and correction for selected
scenarios
o have systems that learn from historic information and behaviour
End-Users cannot build their own services and policies without
becoming technical experts and they must perform manual maintenance
Li, et al. Expires January 8, 2020 [Page 6]
Internet-Draft Intent Classification July 2019
actions. Application developers and end-users/subscribers want to be
able to:
o build their own network services with their own policies via
simple interfaces, without becoming networking experts
o have their network services up and running based on intent and
automation only, without any manual actions or maintenance
3.4. Intent Types that need to be supported
The following intent types need to be supported, in order to address
the requirements from different solutions and intent users:
o Customer network service intent
o for customer self-service
o for service operator orders
o for intent driven network configuration, verification,
correction and optimization
o Network resource management
o For network configuration
o For automated lifecycle management of network configurations
o For network resources (switches, routers, routing, policies,
underlay)
o Cloud and cloud resource management
o For DC configuration, VMs, DB Servers, APP Servers
o For communication between VMs
o For cloud resource lifecycle management (policy driven self-
configuration & auto-scaling & recovery/optimization)
o Network Policy intent
o For security, QoS, application policies, traffic steering, etc
o For configuring & monitoring policies, alarms generation for
non-compliance, auto-recovery
Li, et al. Expires January 8, 2020 [Page 7]
Internet-Draft Intent Classification July 2019
o Task based intents
o For network migration
o For server replacements
o For device replacements
o For network software upgrades
o To automate any tasks that operators/administrator often
perform
o System policies intents
o For intent management system policies
o For design models and policies for network service design
o For design models and policies for network design
o For design workflows, models and policies for task based
intents
o Intents that affect other intents
o It may be task based intent that modifies many other intents.
o The task itself is short-lived, but the modification of other
intents has an impact on their lifecycle, so those changes
must continue to be continuously monitored and self-
corrected/self-optimized.
4. Functional Characteristics and Behavior
Intent can be used to operate immediately on a target (much like
issuing a command), or whenever it is appropriate (e.g., in response
to an event). In either case, intent has a number of behaviors that
serve to further organize its purpose, as described by the following
subsections.
4.1. Persistence
Intents can be classified into transient/persistent intents:
Li, et al. Expires January 8, 2020 [Page 8]
Internet-Draft Intent Classification July 2019
o If intent is transient, it has no lifecycle management. As soon
as the specified operation is successfully carried out, the
intent is finished, and can no longer affect the target object.
o If the intent is persistent, it has lifecycle management. Once
the intent is successfully activated and deployed, the system
will keep all relevant intents active until they are deactivated
or removed.
4.2. Granularity
Intents can have different granularities: high granularity, low
granularity and anything in between.
High granularity intents are more complex to design but are the most
valuable. Intent translation, intent conflict resolution and intent
verification are very complex and require advanced algorithms.
Examples: e2e network service, like customer network service over
physical & virtual network, over access, metro, dc and wan with all
related QoS, security and application policies.
Low granularity intents, like some path checks (can A talk to B) or
individual network service/network/application/user policies, are
the least complex. Their intent translation, intent conflict
resolution and intent verification are much simpler than for high
granularity intents.
Granularity requirements of intents for different users - from the
high granularity e2e network service (e.g. customer network service
over physical/virtual network infrastructure, AN and WAN with all
the QoS/Security/App Policies) to some low granularity path checks.
4.3. Hierarchy
In different phases of the autonomous driving network, the intents
are different. A typical example of autonomous driving network Level
0 to 5 are listed as below.
o Level 0 - Traditional manual network: O&M personnel manually
control the network and obtain network alarms and logs.
o Level 1- Partially automated network: Automated scripts are used
to automate service provisioning, network deployment, and
maintenance. Shallow perception of network status and decision
making suggestions of machine;
Li, et al. Expires January 8, 2020 [Page 9]
Internet-Draft Intent Classification July 2019
o Level 2- Automated network: Automation of most service
provisioning, network deployment, and maintenance Comprehensive
perception of network status and local machine decision making;
o Level 3- Self-optimization network: Deep awareness of network
status and automatic network control, meeting users' network
intentions
o Level 4- Partial autonomous network: In a limited environment,
people do not need to participate in decision-making and adjust
themselves.
o Level 5- Autonomous network: In different network environments
and network conditions, the network can automatically adapt to
and adjust to meet people's intentions.
4.4. Abstracting Intent Operation
The modeling of Policies can be abstracting using the following
three-tuple:
{Context, Capabilities, Constraints}
Context grounds the policy, and determines if it is relevant or not
for the current situation. Capabilities describe the functionality
that the policy can perform. Capabilities take different forms,
depending on the expressivity of the policy as well as the
programming paradigm(s) used. Constraints define any restictions on
the capabilities to be used for that particular context. Metadata
can be optionally attached to each of the elements of the three-
tuple, and may be used to describe how the policy should be used and
how it operates, as well as prescribe any operational dependencies
that must be taken into account. Put another way:
o Context selects policies based on applicability
o Capabilities describe the functionality provided by the policy
o Constraints restrict the capabilities offered and/or the behavior
of the policy
Hence, the difference between imperative, declarative, and other
types of policies lies in how the elements of this three-tuple are
used according to that particular programming paradigm. This is how
[SUPA] was designed: a Policy is a container that aggregates a set
of tatements.
Li, et al. Expires January 8, 2020 [Page 10]
Internet-Draft Intent Classification July 2019
4.5. Policy Subjects and Policy Targets
Policy subject is the actor that performs the action specified in
the policy. It can be the intent management system which executes
the policy. Policy target is a set of managed objects which may be
affected in the policy enforcement.
4.6. Policy Scope
Policies used to manage the behavior of objects that they are
applied to (e.g., the target of the policy). It is useful to
differentiate between the following categories of targets:
o Policies defined for the Customer or End-User
o Policies defined for the management system to act on objects in
the domain that the management system controls
o Policies defined for the management system to act on objects in
one or more domains that the management system does not directly
control
The different origins and views of these three categories of actors
lead to the following important differences:
o Network Knowledge. This area is explored using three exemplary
actors that have different knowledge of the network:
o Customers and end-users do not necessarily know the functional
and operational details of the network that they are using.
Furthermore, most of the actors in this category lack skills
to understand such details; in fact, such knowledge is
typically not relevant to their job. In addition, the network
may not expose these details to its users. This class of
actor focuses on the applications that they run, and uses
services offered by the network. Hence, they want to specify
policies that provide consistent behavior according to their
business needs. They do not have to worry about how the
policies are deployed onto the underlying network, and
especially, whether the policies need to be translated to
different forms to enable network elements to understand
them.
Li, et al. Expires January 8, 2020 [Page 11]
Internet-Draft Intent Classification July 2019
o Application developers work in a set of abstractions defined
by their application and programming environment(s). For
example, many application developers think in terms of
objects (e.g., a VPN). While this makes sense to the
application developer, most network devices do not have a VPN
object per se; rather, the VPN is formed through a set of
configuration statements for that device in concert with
configuration statements for the other devices that
together make up the VPN. Hence, the view of application
developers matches the services provided by the network,
but may not directly correspond to other views of other
actors.
o Management personnel, such as network Administrators, may have
the knowledge of the underlying network. However, they may
not understand the details of the applications and services
of Customers and End-Users.
o Automation. Theoricaly, intents from both end-user and management
system can be automated. In practice, most intents from end-user
are created manually according to business request. End-users do
not create or alter intents unless there is change in business.
Intents from management systems can be created or altered to
reflect with network policy change. For example, end-users create
intents to set up paths between hosts, while the management
system creates an intent to set a global link utilization limit.
5. The Policy Continuum
The Policy Continuum defines the set of actors that will create,
read, use, and manage policy. Each set of actors has their own
terminology and concepts that they are familiar with. This captures
the fact that business people do not want to use CLI, and network
operations center personnel do not want to use non-technical
languages.
6. Involvement of intent in the application of AI to Network Manage
ment
In the application of AI to NM, an intent is expected to be, on the
one hand, a formal definitions of a goal or policy instructed to the
decision system and, on the other hand, a formal definition of the
specific actions that some network controller must perform. Goal
intents and policy intents have different meanings. The former will
establish an objective for the automated management system to
accomplish, such as "avoiding latency to be higher than 10 ms".
Meanwhile, policy intents set the overall regulations and possible
Li, et al. Expires January 8, 2020 [Page 12]
Internet-Draft Intent Classification July 2019
actions that the AI system can use to achieve those goals. Both goal
and policy intents are expected to be provided by humans, although
they must be in some very formal language that can be easily
understood by computers. All those relations make the degree of
formality an important dimension to classify intents so that users,
which here are AI-based agents, can be able to choose the proper
solution to consume them.
To enforce the resulting actions determined by AI-based control
modules, action intents will have a format that avoids
misconceptions as much as possible. This means that they will be
closer to machine language structures than natural (human) language
structures. This can sacrificing some degree of human
understandability, so it forms another dimension in the
classification of intents. This dimension allows automated systems
to discern which format of intent to use in relation to the
possibility and degree of humans to be involved in their exchanges.
Finally, as intents can use different words and languages to refer
to the same concepts, all intents related to AI will be required to
follow a specific ontology. This way, input intents will be easily
semantically translated to formal structures. Output intents will
also be composed by following the ontology, so receivers of those
intents will be able to easily understand them.
7. Security Considerations
This document does not have any Security Considerations.
8. IANA Considerations
This document has no actions for IANA.
9. Contributors
The following people all contributed to creating this document,
listed in alphabetical order:
Richard Meade, Huawei
Weiping Xu, Huawei
10. Acknowledgments
This document has benefited from reviews, suggestions, comments and
proposed text provided by the following members, listed in
Li, et al. Expires January 8, 2020 [Page 13]
Internet-Draft Intent Classification July 2019
alphabetical order: Brian E Carpenter, Juergen Schoenwaelder,
Laurent Ciavaglia, Xiaolin Song.
11. References
11.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC7575] Behringer, M., Pritikin, M., Bjarnason, S., Clemm, A.,
Carpenter, B., Jiang, S., and L. Ciavaglia, "Autonomic
Networking: Definitions and Design Goals", RFC 7575, June
2015.
[RFC8328] Liu, W., Xie, C., Strassner, J., Karagiannis, G., Klyus,
M., Bi, J., Cheng, Y., and D. Zhang, "Policy-Based
Management Framework for the Simplified Use of Policy
Abstractions (SUPA)", March 2018.
[RFC3198] Westerinen, A., Schnizlein, J., Strassner, J.,
Scherling, M., Quinn, B., Herzog, S., Huynh, A., Carlson,
M., Perry, J., Waldbusser, S., "Terminology for Intent-
driven Management", RFC 3198, November 2001.
11.2. Informative References
[RFC6020] Bjorklund, M., "YANG - A Data Modeling Language for the
Network Configuration Protocol (NETCONF)", RFC 6020,
October 2010.
[RFC7285] R. Alimi, R. Penno, Y. Yang, S. Kiesel, S. Previdi, W.
Roome, S. Shalunov, R. Woundy "Application-Layer Traffic
Optimization (ALTO) Protocol", September 2014.
[ANIMA] Du, Z., "ANIMA Intent Policy and Format", 2017,
.
[ONF] ONF, "Intent Definition Principles", 2017,
.
[ONOS] ONOS, "ONOS Intent Framework", 2017,
.
Li, et al. Expires January 8, 2020 [Page 14]
Internet-Draft Intent Classification July 2019
[SUPA] Strassner, J., "Simplified Use of Policy Abstractions",
2017, .
[ANIMA-Prefix] Jiang, S., Du, Z., Carpenter, B., and Q. Sun,
"Autonomic IPv6 Edge Prefix Management in Large-scale
Networks", draft-ietf-anima-prefix-management-07 (work in
progress), December 2017.
Li, et al. Expires January 8, 2020 [Page 15]
Internet-Draft Intent Classification July 2019
Authors' Addresses
Chen Li
China Telecom
No.118 Xizhimennei street, Xicheng District
Beijing 100035
P.R. China
Email: lichen.bri@chinatelecom.cn
Ying Cheng
China Unicom
No.21 Financial Street, XiCheng District
Beijing 100033
P.R. China
Email: chengying10@chinaunicom.cn
John Strassner
Huawei Technologies
2330 Central Expressway
Santa Clara, CA 95138
United States of America
Email: john.sc.strassner@huawei.com
Olga Havel
Huawei Technologies
Email: olga.havel@huawei.com
Will(Shucheng) Liu
Huawei Technologies
Bantian, Longgang District
Shenzhen 518129
P.R. China
Email: liushucheng@huawei.com
Pedro Martinez-Julia
NICT
Japan
Email: pedro@nict.go.jp
Jeferson Campos Nobre
University of Vale do Rio dos Sinos
Porto Alegre
Brazil
Email: jcnobre@inf.ufrgs.br
Diego R. Lopez
Telefonica I+D
Li, et al. Expires January 8, 2020 [Page 16]
Internet-Draft Intent Classification July 2019
Don Ramon de la Cruz, 82
Madrid 28006
Spain
Email: diego.r.lopez@telefonica.com
Li, et al. Expires January 8, 2020 [Page 17]