Re: [OPSAWG] AD review of draft-ietf-opsawg-ntf-07

["Rob Wilton (rwilton)" <[e-mail]>]

Sorry, the IETF mail filters have truncated my email.  I'll resend as HTML to see if that helps.

Rob


-----Original Message-----
From: Rob Wilton (rwilton) <[e-mail]> 
Sent: 06 October 2021 11:35
To: [e-mail]
Cc: [e-mail]
Subject: AD review of draft-ietf-opsawg-ntf-07

Hi,

Here is my belated AD review of draft-ietf-opsawg-ntf-07.txt.

I would like to thank you for the effort that you have put into this document, and apologise for my long delay in reviewing it.

Broadly, I think that this is a good and useful framework, but in some of the latter parts of the document it seems to give prominence to protocols that I don't think have IETF consensus behind them yet (particularly DNP).  I have flagged specific comments in comments inline within the document, but I think that the document will have been accuracy/longevity if text about the potential technologies is mostly kept to the appendices.

There were quite a lot of cases where the text doesn't scan, or read easily, particularly in the latter sections of this document, although I acknowledge that none of the authors appear to be native English speakers.  Ideally, these sorts of issues would have been highlighted and addressed during WG LC.  Although the RFC editor will improve the language of the documents, making the improvements now before IESG review will aid its passage, and hopefully result in a better document when it is published.  I have flagged and proposed alternative text/grammar where possible.  Once you have made the markups and resolved the issues/questions that I have raised then I can run it through a grammar checking tool (Lar's will run an equivalent tool during IESG review anyway ...)

All of my comments are directly inline, please search for "RW" or "RW:"





OPSAWG                                                           H. Song
Internet-Draft                                                 Futurewei
Intended status: Informational                                    F. Qin
Expires: August 23, 2021                                    China Mobile
                                                       P. Martinez-Julia
                                                                    NICT
                                                            L. Ciavaglia
                                                                   Nokia
                                                                 A. Wang
                                                           China Telecom
                                                       February 19, 2021


                      Network Telemetry Framework
                        draft-ietf-opsawg-ntf-07

Abstract

   Network telemetry is a technology for gaining network insight and
   facilitating efficient and automated network management.  It
   encompasses various techniques for remote data generation,
   collection, correlation, and consumption.  This document describes an
   architectural framework for network telemetry, motivated by
   challenges that are encountered as part of the operation of networks
   and by the requirements that ensue.  Network telemetry, as
   necessitated by best industry practices, covers technologies and
   protocols that extend beyond conventional network Operations,
   
   Administration, and Management (OAM).  The presented network
   telemetry framework promises flexibility, scalability, accuracy,
   coverage, and performance.  In addition, it facilitates the
   implementation of automated control loops to address both today's and
   tomorrow's network operational needs.  This document clarifies the
   terminologies and classifies the modules and components of a network
   telemetry system from several different perspectives.  The framework
   and taxonomy help to set a common ground for the collection of
   related work and provide guidance for related technique and standard
   developments.

RW:
I would suggest condensing the abstract to the following and move the other text to the introduction if it is not already covered there.

   Network telemetry is a technology for gaining network insight and
   facilitating efficient and automated network management.  It
   encompasses various techniques for remote data generation,
   collection, correlation, and consumption.  This document describes an
   architectural framework for network telemetry, motivated by
   challenges that are encountered as part of the operation of networks
   and by the requirements that ensue.  This document clarifies the
   terminologies and classifies the modules and components of a network
   telemetry system from several different perspectives.  The framework
   and taxonomy help to set a common ground for the collection of
   related work and provide guidance for related technique and standard
   developments.


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).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at https://datatracker.ietf.org/drafts/current/.




Song, et al.             Expires August 23, 2021                [Page 1]


Internet-Draft         Network Telemetry Framework         February 2021


   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 August 23, 2021.

Copyright Notice

   Copyright (c) 2021 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
   (https://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.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   3
   2.  Glossary  . . . . . . . . . . . . . . . . . . . . . . . . . .   4
   3.  Background  . . . . . . . . . . . . . . . . . . . . . . . . .   6
     3.1.  Telemetry Data Coverage . . . . . . . . . . . . . . . . .   7
     3.2.  Use Cases . . . . . . . . . . . . . . . . . . . . . . . .   7
     3.3.  Challenges  . . . . . . . . . . . . . . . . . . . . . . .   9
     3.4.  Network Telemetry . . . . . . . . . . . . . . . . . . . .  10
   4.  The Necessity of a Network Telemetry Framework  . . . . . . .  12
   5.  Network Telemetry Framework . . . . . . . . . . . . . . . . .  13
     5.1.  Top Level Modules . . . . . . . . . . . . . . . . . . . .  14
       5.1.1.  Management Plane Telemetry  . . . . . . . . . . . . .  17
       5.1.2.  Control Plane Telemetry . . . . . . . . . . . . . . .  17
       5.1.3.  Forwarding Plane Telemetry  . . . . . . . . . . . . .  18
       5.1.4.  External Data Telemetry . . . . . . . . . . . . . . .  20
     5.2.  Second Level Function Components  . . . . . . . . . . . .  21
     5.3.  Data Acquisition Mechanism and Type Abstraction . . . . .  22
     5.4.  Mapping Existing Mechanisms into the Framework  . . . . .  24
   6.  Evolution of Network Telemetry Applications . . . . . . . . .  25
   7.  Security Considerations . . . . . . . . . . . . . . . . . . .  26
   8.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  27
   9.  Contributors  . . . . . . . . . . . . . . . . . . . . . . . .  27
   10. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . .  28
   11. Informative References  . . . . . . . . . . . . . . . . . . .  28
   Appendix A.  A Survey on Existing Network Telemetry Techniques  .  32



Song, et al.             Expires August 23, 2021                [Page 2]


Internet-Draft         Network Telemetry Framework         February 2021


     A.1.  Management Plane Telemetry  . . . . . . . . . . . . . . .  32
       A.1.1.  Push Extensions for NETCONF . . . . . . . . . . . . .  32
       A.1.2.  gRPC Network Management Interface . . . . . . . . . .  32
     A.2.  Control Plane Telemetry . . . . . . . . . . . . . . . . .  33
       A.2.1.  BGP Monitoring Protocol . . . . . . . . . . . . . . .  33
     A.3.  Data Plane Telemetry  . . . . . . . . . . . . . . . . . .  33
       A.3.1.  The Alternate Marking (AM) technology . . . . . . . .  33
       A.3.2.  Dynamic Network Probe . . . . . . . . . . . . . . . .  34
       A.3.3.  IP Flow Information Export (IPFIX) protocol . . . . .  35
       A.3.4.  In-Situ OAM . . . . . . . . . . . . . . . . . . . . .  35
       A.3.5.  Postcard Based Telemetry  . . . . . . . . . . . . . .  35
     A.4.  External Data and Event Telemetry . . . . . . . . . . . .  35
       A.4.1.  Sources of External Events  . . . . . . . . . . . . .  36
       A.4.2.  Connectors and Interfaces . . . . . . . . . . . . . .  37
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  37

1.  Introduction

   Network visibility is the ability of management tools to see the
   state and behavior of a network, which is essential for successful
   network operation.  Network Telemetry revolves around network data
   that can help provide insights about the current state of the
   network, including network devices, forwarding, control, and
   management planes, and that can be generated and obtained through a
   variety of techniques, including but not limited to network
   instrumentation and measurements, and that can be processed for
   purposes ranging from service assurance to network security using a
   wide variety of techniques including machine learning, data analysis,
   and correlation.  In this document, Network Telemetry refer to both
   the data itself (i.e., "Network Telemetry Data"), and the techniques
   and processes used to generate, export, collect, and consume that
   data for use by potentially automated management applications.
   Network telemetry extends beyond the conventional network Operations,
   Administration, and Management (OAM) techniques and expects to
   support better flexibility, scalability, accuracy, coverage, and
   performance.
   
RW: I suggest 'historical' rather than 'conventional'


   However, the term of network telemetry lacks a solid and unambiguous
   definition.  The scope and coverage of it cause confusion and
   misunderstandings.  It is beneficial to clarify the concept and
   provide a clear architectural framework for network telemetry, so we
   can articulate the technical field, and better align the related
   techniques and standard works.

RW: Rather than term of, perhaps 'the term "network telemetry" lacks an
    unambiguous definition'.


   To fulfill such an undertaking, we first discuss some key
   characteristics of network telemetry which set a clear distinction
   from the conventional network OAM and show that some conventional OAM
   technologies can be considered a subset of the network telemetry



Song, et al.             Expires August 23, 2021                [Page 3]


Internet-Draft         Network Telemetry Framework         February 2021


   technologies.  We then provide an architectural framework for network
   telemetry which includes four modules, each concerned with a
   different category of telemetry data and corresponding procedures.
   All the modules are internally structured in the same way, including
   components that allow to configure data sources with regards to what
   data to generate and how to make that available to client
   applications, components that instrument the underlying data sources,
   and components that perform the actual rendering, encoding, and
   exporting of the generated data.  We show how the network telemetry
   framework can benefit the current and future network operations.
   Based on the distinction of modules and function components, we can
   map the existing and emerging techniques and protocols into the
   framework.  The framework can also simplify the tasks for designing,
   maintaining, and understanding a network telemetry system.  At last,
   we outline the evolution stages of the network telemetry system and
   discuss the potential security concerns.

   The purpose of the framework and taxonomy is to set a common ground
   for the collection of related work and provide guidance for future
   technique and standard developments.  To the best of our knowledge,
   this document is the first such effort for network telemetry in
   industry standards organizations.


2.  Glossary

   Before further discussion, we list some key terminology and acronyms
   used in this documents.  We make an intended differentiation between
   the terms of network telemetry and OAM.  However, it should be
   understood that there is not a hard-line distinction between the two
   concepts.  Rather, network telemetry is considered as the extension
   of OAM.  It covers all the existing OAM protocols but puts more
   emphasis on the newer and emerging techniques and protocols
   concerning all aspects of network data from acquisition to
   consumption.


RW: 
Nit: "this documents." -> "this document."
Nit: "as an extension" rather than "as the extension".

   AI:  Artificial Intelligence.  In network domain, AI refers to the
      machine-learning based technologies for automated network
      operation and other tasks.

   AM:  Alternate Marking, a flow performance measurement method,
      specified in [RFC8321].

   BMP:  BGP Monitoring Protocol, specified in [RFC7854].

   DNP:  Dynamic Network Probe, referring to programmable in-network
      sensors for network monitoring and measurement.





Song, et al.             Expires August 23, 2021                [Page 4]


Internet-Draft         Network Telemetry Framework         February 2021


   DPI:  Deep Packet Inspection, referring to the techniques that
      examines packet beyond packet L3/L4 headers.

   gNMI:  gRPC Network Management Interface, a network management
      protocol from OpenConfig Operator Working Group, mainly
      contributed by Google.  See [gnmi] for details.

   gRPC:  gRPC Remote Procedure Call, a open source high performance RPC
      framework that gNMI is based on.  See [grpc] for details.

   IPFIX:  IP Flow Information Export Protocol, specified in [RFC7011].

   IOAM:  In-situ OAM, a dataplane on-path telemetry technique.

   NETCONF:  Network Configuration Protocol, specified in [RFC6241].

   NetFlow:  A Cisco protocol for flow record collecting, described in
      [RFC3594].

   Network Telemetry:  The process and instrumentation for acquiring and
      utilizing network data remotely for network monitoring and
      operation.  A general term for a large set of network visibility
      techniques and protocols, concerning aspects like data generation,
      collection, correlation, and consumption.  Network telemetry
      addresses the current network operation issues and enables smooth
      evolution toward future intent-driven autonomous networks.

   NMS:  Network Management System, referring to applications that allow
      network administrators manage a network.

RW: referring to => refers to applications that allow network administrators to manage a network.



   OAM:  Operations, Administration, and Maintenance.  A group of
      network management functions that provide network fault
      indication, fault localization, performance information, and data
      and diagnosis functions.  Most conventional network monitoring
      techniques and protocols belong to network OAM.

   PBT:  Postcard-Based Telemetry, a dataplane on-path telemetry
      technique.
      
   SMIv2  Structure of Management Information Version 2, specified in
      [RFC2578].

RW:
Is SMIv2 a better reference than MIBs, that readers are more likely to be familiar with?


   SNMP:  Simple Network Management Protocol.  Version 1 and 2 are
      specified in [RFC1157] and [RFC3416], respectively.

   YANG:  The abbreviation of "Yet Another Next Generation".  YANG is a
      data modeling language for the definition of data sent over

RW:
Nit: Please drop the first sentence, and add a reference to RFC 7950.



Song, et al.             Expires August 23, 2021                [Page 5]


Internet-Draft         Network Telemetry Framework         February 2021


      network management protocols such as the NETCONF and RESTCONF.
      YANG is defined in [RFC6020].

   YANG ECA  A YANG model for Event-Condition-Action policies, defined
      in [I-D.wwx-netmod-event-yang].

   YANG PUSH:  A method to subscribe pushed data from remote YANG
      datastore on network devices.  Details are specified in [RFC8641]
      and [RFC8639].
      
RW:
Perhaps borrow from the abstract in RFC 8641.
  "A mechanism that allows subscriber applications to request a
   stream of updates from a YANG datastore on a network device".  Details are ...      


3.  Background

   The term "big data" is used to describe the extremely large volume of
   data sets that can be analyzed computationally to reveal patterns,
   trends, and associations.  Networks are undoubtedly a source of big
   data because of their scale and the volume of network traffic they
   forward.  It is easy to see that network operations can benefit from
   network big data.
   
RW:
Also need to consider privacy.

I think that we need to be careful not to imply that the intention here is to read/snoop on the data being carried over the network rather than gather insights into flows 

    

   Today one can access advanced big data analytics capability through a
   plethora of commercial and open source platforms (e.g., Apache
   Hadoop), tools (e.g., Apache Spark), and techniques (e.g., machine
   learning).  Thanks to the advance of computing and storage
   technologies, network big data analytics gives network operators an
   opportunity to gain network insights and move towards network
   autonomy.  Some operators start to explore the application of
   Artificial Intelligence (AI) to make sense of network data.  Software
   tools can use the network data to detect and react on network faults,
   anomalies, and policy violations, as well as predicting future
   events.  In turn, the network policy updates for planning, intrusion
   prevention, optimization, and self-healing may be applied.

   It is conceivable that an autonomic network [RFC7575] is the logical
   next step for network evolution following Software Defined Network
   (SDN), aiming to reduce (or even eliminate) human labor, make more
   efficient use of network resources, and provide better services more
   aligned with customer requirements.  Intent-based Networking (IBN)
   [I-D.irtf-nmrg-ibn-concepts-definitions] requires network visibility
   and telemetry data in order to ensure that the network is behaving as
   intended.  Although it takes time to reach the ultimate goal, the
   journey has started nevertheless.
 
RW:
It would be helpful for the text to link autonomic networking and Intent based networking, perhaps:
The related technique of Intent-based Networking [...] requires ...

RW:
Not sure that the last sentence of the paragraph is required.
 
   

   However, while the data processing capability is improved and
   applications are hungry for more data, the networks lag behind in
   extracting and translating network data into useful and actionable
   information in efficient ways.  The system bottleneck is shifting
   from data consumption to data supply.  Both the number of network
   nodes and the traffic bandwidth keep increasing at a fast pace.  The



Song, et al.             Expires August 23, 2021                [Page 6]


Internet-Draft         Network Telemetry Framework         February 2021


   network configuration and policy change at smaller time slots than
   before.  More subtle events and fine-grained data through all network
   planes need to be captured and exported in real time.  In a nutshell,
   it is a challenge to get enough high-quality data out of the network
   in a manner that is efficient, timely, and flexible.  Therefore, we
   need to survey the existing technologies and protocols and identify
   any potential gaps.

   In the remainder of this section, first we clarify the scope of
   network data (i.e., telemetry data) concerned in the context.  Then,
   we discuss several key use cases for today's and future network
   operations.  Next, we show why the current network OAM techniques and
   protocols are insufficient for these use cases.  The discussion
   underlines the need of new methods, techniques, and protocols which
   we assign under the umbrella term - Network Telemetry.

RW:
We should also include the possibilty of extending existing protocols, methods, techniques.


3.1.  Telemetry Data Coverage

   Any information that can be extracted from networks (including data
   plane, control plane, and management plane) and used to gain
   visibility or as basis for actions is considered telemetry data.  It
   includes statistics, event records and logs, snapshots of state,
   configuration data, etc.  It also covers the outputs of any active
   and passive measurements [RFC7799].  Specially, raw data can be
   processed in-network before being sent to a data consumer.  Such
   processed data is also considered telemetry data.  A classification
   of telemetry data is provided in Section 5.

RW:
Specially - I would expand this.  Perhaps: "In some cases, raw data is processed before being sent .."
We should also discuss the quality of data, i.e., less, higher quality data may be better than lots of low quality data.


3.2.  Use Cases

   The following set of use cases is essential for network operations.
   While the list is by no means exhaustive, it is enough to highlight
   the requirements for data velocity, variety, volume, and veracity in
   networks.

   o  Security: Network intrusion detection and prevention systems need
      to monitor network traffic and activities and act upon anomalies.
      Given increasingly sophisticated attack vector coupled with
      increasingly severe consequences of security breaches, new tools
      and techniques need to be developed, relying on wider and deeper
      visibility into networks.
      
RW:
I agree with this, but it might be good to emphasize that the goal is
to get to a place where this can be done without any, or only minimal,
human intervention.


   o  Policy and Intent Compliance: Network policies are the rules that
      constraint the services for network access, provide service
      differentiation, or enforce specific treatment on the traffic.
      For example, a service function chain is a policy that requires
      the selected flows to pass through a set of ordered network
      functions.  Intent, as defined in

RW:
constraint => constrain


Song, et al.             Expires August 23, 2021                [Page 7]


Internet-Draft         Network Telemetry Framework         February 2021


      [I-D.irtf-nmrg-ibn-concepts-definitions], is a set of operational
      goal that a network should meet and outcomes that a network is
      supposed to deliver, defined in a declarative manner without
      specifying how to achieve or implement them.  An intent requires a
      complex translation and mapping process before being applied on
      networks.  While a policy or an intent is enforced, the compliance
      needs to be verified and monitored continuously, relying on
      visibility that is provided through network telemetry data, and
      any violation needs to be reported immediately.

RW: 
Does it not also rely on visibility of the network to potentially modify
the mapping to ensure that the intent remains in force?

   o  SLA Compliance: A Service-Level Agreement (SLA) defines the level
      of service a user expects from a network operator, which include
      the metrics for the service measurement and remedy/penalty
      procedures when the service level misses the agreement.  Users
      need to check if they get the service as promised and network
      operators need to evaluate how they can deliver the services that
      can meet the SLA based on realtime network telemetry data,
      including data from network measurements.

   o  Root Cause Analysis: Any network failure can be the effect of a
      sequence of chained events.  Troubleshooting and recovery require
      quick identification of the root cause of any observable issues.
      However, the root cause is not always straightforward to identify,
      especially when the failure is sporadic and the number of event
      messages, both related and unrelated to the same cause, is
      overwhelming.  While machine learning technologies can be used for
      root cause analysis, it up to the network to sense and provide the
      relevant data to feed into machine learning applications.

RW:
In these sorts of scenarios, I would expect additional detailed diagnostics information to be requested from the device to figure out the root cause.  Or specifically, I think that this would contain data that wouldn't normally be exported via telemetry.


   o  Network Optimization: This covers all short-term and long-term
      network optimization techniques, including load balancing, Traffic
      Engineering (TE), and network planning.  Network operators are
      motivated to optimize their network utilization and differentiate
      services for better Return On Investment (ROI) or lower Capital
      Expenditures (CAPEX).  The first step is to know the real-time
      network conditions before applying policies for traffic
      manipulation.  In some cases, micro-bursts need to be detected in
      a very short time-frame so that fine-grained traffic control can
      be applied to avoid network congestion.  Long-term planning of
      network capacity and topology requires analysis of real-world
      network telemetry data that is obtained over long periods of time.

   o  Event Tracking and Prediction: The visibility into traffic path
      and performance is critical for services and applications that
      rely on healthy network operation.  Numerous related network
      events are of interest to network operators.  For example, Network
      operators want to learn where and why packets are dropped for an
      application flow.  They also want to be warned of issues in



Song, et al.             Expires August 23, 2021                [Page 8]


Internet-Draft         Network Telemetry Framework         February 2021


      advance so proactive actions can be taken to avoid catastrophic
      consequences.

3.3.  Challenges

   For a long time, network operators have relied upon SNMP [RFC3416],
   Command-Line Interface (CLI), or Syslog to monitor the network.  Some
   other OAM techniques as described in [RFC7276] are also used to
   facilitate network troubleshooting.  These conventional techniques
   are not sufficient to support the above use cases for the following
   reasons:

   o  Most use cases need to continuously monitor the network and
      dynamically refine the data collection in real-time.  The poll-
      based low-frequency data collection is ill-suited for these
      applications.  Subscription-based streaming data directly pushed
      from the data source (e.g., the forwarding chip) is preferred to
      provide enough data quantity and precision at scale.

   o  Comprehensive data is needed from packet processing engine to
      traffic manager, from line cards to main control board, from user
      flows to control protocol packets, from device configurations to
      operations, and from physical layer to application layer.
      Conventional OAM only covers a narrow range of data (e.g., SNMP
      only handles data from the Management Information Base (MIB)).
      Traditional network devices cannot provide all the necessary
      probes.  More open and programmable network devices are therefore
      needed.

   o  Many application scenarios need to correlate network-wide data
      from multiple sources (i.e., from distributed network devices,
      different components of a network device, or different network
      planes).  A piecemeal solution is often lacking the capability to
      consolidate the data from multiple sources.  The composition of a
      complete solution, as partly proposed by Autonomic Resource
      Control Architecture(ARCA)
      [I-D.pedro-nmrg-anticipated-adaptation], will be empowered and
      guided by a comprehensive framework.

   o  Some of the conventional OAM techniques (e.g., CLI and Syslog)
      lack a formal data model.  The unstructured data hinder the tool
      automation and application extensibility.  Standardized data
      models are essential to support the programmable networks.

   o  Although some conventional OAM techniques support data push (e.g.,
      SNMP Trap [RFC2981][RFC3877], Syslog, and sFlow), the pushed data
      are limited to only predefined management plane warnings (e.g.,
      SNMP Trap) or sampled user packets (e.g., sFlow).  Network



Song, et al.             Expires August 23, 2021                [Page 9]


Internet-Draft         Network Telemetry Framework         February 2021


      operators require the data with arbitrary source, granularity, and
      precision which are beyond the capability of the existing
      techniques.

   o  The conventional passive measurement techniques can either consume
      excessive network resources and render excessive redundant data,
      or lead to inaccurate results; on the other hand, the conventional
      active measurement techniques can interfere with the user traffic
      and their results are indirect.  Techniques that can collect
      direct and on-demand data from user traffic are more favorable.

   These challenges were addressed by newer standards and techniques
   (e.g., IPFIX/Netflow, PSAMP, IOAM, and YANG-Push) and more are
   emerging.  These standards and techniques need to be recognized and
   accommodated in a new framework.

3.4.  Network Telemetry

   Network telemetry has emerged as a mainstream technical term to refer
   to the network data collection and consumption techniques.  Several
   network telemetry techniques and protocols (e.g., IPFIX [RFC7011] and
   gRPC [grpc]) have been widely deployed.  Network telemetry allows
   separate entities to acquire data from network devices so that data
   can be visualized and analyzed to support network monitoring and
   operation.  Network telemetry covers the conventional network OAM and
   has a wider scope.  It is expected that network telemetry can provide
   the necessary network insight for autonomous networks and address the
   shortcomings of conventional OAM techniques.

   Network telemetry usually assumes machines as data consumers rather
   than human operators.  Hence, the network telemetry can directly
   trigger the automated network operation, while in contrast some
   conventional OAM tools are designed and used to help human operators
   to monitor and diagnose the networks and guide manual network
   operations.  Such a proposition leads to very different techniques.

   Although new network telemetry techniques are emerging and subject to
   continuous evolution, several characteristics of network telemetry
   have been well accepted.  Note that network telemetry is intended to
   be an umbrella term covering a wide spectrum of techniques, so the
   following characteristics are not expected to be held by every
   specific technique.

   o  Push and Streaming: Instead of polling data from network devices,
      telemetry collectors subscribe to streaming data pushed from data
      sources in network devices.





Song, et al.             Expires August 23, 2021               [Page 10]


Internet-Draft         Network Telemetry Framework         February 2021


   o  Volume and Velocity: The telemetry data is intended to be consumed
      by machines rather than by human being.  Therefore, the data
      volume can be huge and the processing is optimized for the needs
      of automation in realtime.

   o  Normalization and Unification: Telemetry aims to address the
      overall network automation needs.  Efforts are made to normalize
      the data representation and unify the protocols, so to simplify
      data analysis and provide integrated analysis across heterogeneous
      devices and data sources across a network.

   o  Model-based: The telemetry data is modeled in advance which allows
      applications to configure and consume data with ease.

   o  Data Fusion: The data for a single application can come from
      multiple data sources (e.g., cross-domain, cross-device, and
      cross-layer) and needs to be correlated to take effect.

   o  Dynamic and Interactive: Since the network telemetry means to be
      used in a closed control loop for network automation, it needs to
      run continuously and adapt to the dynamic and interactive queries
      from the network operation controller.

   In addition, an ideal network telemetry solution may also have the
   following features or properties:

   o  In-Network Customization: The data that is generated can be
      customized in network at run-time to cater to the specific need of
      applications.  This needs the support of a programmable data plane
      which allows probes with custom functions to be deployed at
      flexible locations.

   o  In-Network Data Aggregation and Correlation: Network devices and
      aggregation points can work out which events and what data needs
      to be stored, reported, or discarded thus reducing the load on the
      central collection and processing points while still ensuring that
      the right information is ready to be processed in a timely way.

   o  In-Network Processing: Sometimes it is not necessary or feasible
      to gather all information to a central point to be processed and
      acted upon.  It is possible for the data processing to be done in
      network, allowing reactive actions to be taken locally.

   o  Direct Data Plane Export: The data originated from the data plane
      forwarding chips can be directly exported to the data consumer for
      efficiency, especially when the data bandwidth is large and the
      real-time processing is required.




Song, et al.             Expires August 23, 2021               [Page 11]


Internet-Draft         Network Telemetry Framework         February 2021


   o  In-band Data Collection: In addition to the passive and active
      data collection approaches, the new hybrid approach allows to
      directly collect data for any target flow on its entire forwarding
      path [I-D.song-opsawg-ifit-framework].

   It is worth noting that a network telemetry system should not be
   intrusive to normal network operations by avoiding the pitfall of the
   "observer effect".  That is, it should not change the network
   behavior and affect the forwarding performance.  Otherwise, the whole
   purpose of network telemetry is compromised.

   Although in many cases a system for network telemetry involves a
   remote data collecting and consuming entity, it is important to
   understand that there are no inherent assumptions about how a system
   should be architected.  Telemetry data producers and consumers can
   work in distributed or peer-to-peer fashions rather than assuming a
   centralized data consuming entity.  In such cases, a network node can
   be the direct consumer of telemetry data from other nodes.

4.  The Necessity of a Network Telemetry Framework

RW: I think that the structure of the document might be better if this was a section 3.5 of the background rather than it's own top level section?

   Network data analytics and machine-learning technologies are applied
   for network operation automation, relying on abundant and coherent
   data from networks.  Data acquisition that is limited to a single
   source and static in nature will in many cases not be sufficient to
   meet an application's telemetry data needs.  As a result, multiple
   data sources, involving a variety of techniques and standards, will
   need to be integrated.  It is desirable to have a framework that
   classifies and organizes different telemetry data source and types,
   defines different components of a network telemetry system and their
   interactions, and helps coordinate and integrate multiple telemetry
   approaches across layers.  This allows flexible combinations of data
   for different applications, while normalizing and simplifying
   interfaces.  In detail, such a framework would benefit application
   development for the following reasons:

   o  Future networks, autonomous or otherwise, depend on holistic and
      comprehensive network visibility.  All the use cases and
      applications are better to be supported uniformly and coherently
      under a single intelligent agent using an integrated, converged
      mechanism and common telemetry data representations wherever
      feasible.  Therefore, the protocols and mechanisms should be
      consolidated into a minimum yet comprehensive set.  A telemetry
      framework can help to normalize the technique developments.

   o  Network visibility presents multiple viewpoints.  For example, the
      device viewpoint takes the network infrastructure as the
      monitoring object from which the network topology and device



Song, et al.             Expires August 23, 2021               [Page 12]


Internet-Draft         Network Telemetry Framework         February 2021


      status can be acquired; the traffic viewpoint takes the flows or
      packets as the monitoring object from which the traffic quality
      and path can be acquired.  An application may need to switch its
      viewpoint during operation.  It may also need to correlate a
      service and its impact on user experience to acquire the
      comprehensive information.

   o  Applications require network telemetry to be elastic in order to
      make efficient use of network resources and reduce the impact of
      processing related to network telemetry on network performance.
      For example, routine network monitoring should cover the entire
      network with a low data sampling rate.  Only when issues arise or
      critical trends emerge should telemetry data source be modified
      and telemetry data rates boosted as needed.

   o  Efficient data fusion is critical for applications to reduce the
      overall quantity of data and improve the accuracy of analysis.

   A telemetry framework collects together all of the telemetry-related
   works from different sources and working groups within IETF.  This
   makes it possible to assemble a comprehensive network telemetry
   system and to avoid repetitious or redundant work.  The framework
   should cover the concepts and components from the standardization
   perspective.  This document describes the modules which make up a
   network telemetry framework and decomposes the telemetry system into
   a set of distinct components that existing and future work can easily
   map to.

5.  Network Telemetry Framework

   The top level network telemetry framework partitions the network
   telemetry into four modules based on the telemetry data object source
   and represents their relationship.  At the next level, the framework
   decomposes each module into separate components.  Each of the modules
   follows the same underlying structure, with one component dedicated
   to the configuration of data subscriptions and data sources, a second
   component dedicated to encoding and exporting data, and a third
   component instrumenting the generation of telemetry related to the
   underlying resources.  Throughout the framework, the same set of
   abstract data acquiring mechanisms and data types are applied.  The
   two-level architecture with the uniform data abstraction helps
   accurately pinpoint a protocol or technique to its position in a
   network telemetry system or disaggregate a network telemetry system
   into manageable parts.


RW: Relationship of telemetry data vs get requests.  I.e., isn't telemtry just push rather than pulling data.




Song, et al.             Expires August 23, 2021               [Page 13]


Internet-Draft         Network Telemetry Framework         February 2021


5.1.  Top Level Modules

   Telemetry can be applied on the forwarding plane, the control plane,
   and the management plane in a network, as well as other sources out
   of the network, as shown in Figure 1.  Therefore, we categorize the
   network telemetry into four distinct modules with each having its own
   interface to Network Operation Applications.

                   +------------------------------+
                   |                              |
                   |       Network Operation      |<-------+
                   |          Applications        |        |
                   |                              |        |
                   +------------------------------+        |
                        ^      ^           ^               |
                        |      |           |               |
                        V      |           V               V
                   +-----------|---+--------------+  +-----------+
                   |           |   |              |  |           |
                   | Control Pl|ane|              |  | External  |
                   | Telemetry | <--->            |  | Data and  |
                   |           |   |              |  | Event     |
                   |      ^    V   |  Management  |  | Telemetry |
                   +------|--------+  Plane       |  |           |
                   |      V        |  Telemetry   |  +-----------+
                   | Forwarding    |              |
                   | Plane       <--->            |
                   | Telemetry     |              |
                   |               |              |
                   +---------------+--------------+

                Figure 1: Modules in Layer Category of NTF

RW:
In this diagram, for me at least, I think that it would more natural to have Management Plane on the left, and Control/ Forwarding Plane on the right.

   The rationale of this partition lies in the different telemetry data
   objects which result in different data source and export locations.
   Such differences have profound implications on in-network data
   programming and processing capability, data encoding and transport
   protocol, and required data bandwidth and latency.

RW:
Data can be sent directly, or proxied via the control and management planes