NMOP V. Riccobene
Internet-Draft Huawei
Intended status: Experimental T. Graf
Expires: 4 September 2025 W. Du
Swisscom
A. Huang Feng
INSA-Lyon
3 March 2025
An Experiment: Network Anomaly Lifecycle
draft-ietf-nmop-network-anomaly-lifecycle-02
Abstract
Network Anomaly Detection is the act of detecting problems in the
network. Accurately detect problems is very challenging for network
operators in production networks. Good results require a lot of
expertise and knowledge around both the implied network technologies
and the connectivity services provided to customers, apart from a
proper monitoring infrastructure. In order to facilitate network
anomaly detection, novel techniques are being introduced, including
programmatical, rule-based and AI-based, with the promise of
improving scalability and the hope to keep a high detection accuracy.
To guarantee acceptable results, the process needs to be properly
designed, adopting well-defined stages to accurately collect evidence
of anomalies, validate their relevancy and improve the detection
systems over time, iteratively.
This document describes a well-defined approach on managing the
lifecycle process of a network anomaly detection system, spanning
across the recording of its output and its iterative refinement, in
order to facilitate network engineers to interact with the network
anomaly detection system, enable the "human-in-the-loop" paradigm and
refine the detection abilities over time. The major contributions of
this document are: the definition of three key stages of the
lifecycle process, the definition of a state machine for each anomaly
annotation on the system and the definition of YANG data models
describing a comprehensive format for the anomaly labels, allowing a
well-structured exchange of those between all the interested actors.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Riccobene, et al. Expires 4 September 2025 [Page 1]
Internet-Draft network-anomaly-lifecycle March 2025
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/.
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 4 September 2025.
Copyright Notice
Copyright (c) 2025 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 Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Discussion Venues . . . . . . . . . . . . . . . . . . . . . . 3
2. Status of this document . . . . . . . . . . . . . . . . . . . 3
3. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
4. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4
5. Defining Desired States . . . . . . . . . . . . . . . . . . . 5
6. Lifecycle of a Network Anomaly . . . . . . . . . . . . . . . 6
6.1. Network Anomaly Detection . . . . . . . . . . . . . . . . 7
6.2. Network Anomaly Validation . . . . . . . . . . . . . . . 8
6.3. Network Anomaly Refinement . . . . . . . . . . . . . . . 8
7. Introducing a Label Store for Network Anomaly labels . . . . 9
8. Network Anomaly State Machine . . . . . . . . . . . . . . . . 9
9. Network Anomaly Data Model . . . . . . . . . . . . . . . . . 10
9.1. Overview of the Data Model for the Relevant State and all
the related entities . . . . . . . . . . . . . . . . . . 10
10. Implementation status . . . . . . . . . . . . . . . . . . . . 21
10.1. Antagonist . . . . . . . . . . . . . . . . . . . . . . . 21
11. Security Considerations . . . . . . . . . . . . . . . . . . . 21
12. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 21
13. Normative References . . . . . . . . . . . . . . . . . . . . 21
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 23
Riccobene, et al. Expires 4 September 2025 [Page 2]
Internet-Draft network-anomaly-lifecycle March 2025
1. Discussion Venues
This note is to be removed before publishing as an RFC.
Discussion of this document takes place on the Network Management and
Operations Area Working Group Working Group mailing list
(nmop@ietf.org), which is archived at
https://mailarchive.ietf.org/arch/browse/nmop/.
Source for this draft and an issue tracker can be found at
https://github.com/network-analytics/draft-netana-nmop-network-
anomaly-lifecycle.
2. Status of this document
This document is experimental. The main goal of this document is to
propose an iterative lifecycle process to network anomaly detection
by proposing a data model for metadata to be addressed at different
lifecycle stages.
The experiment consists of verifying whether the approach is usable
in real use case scenarios to support proper refinement and
adjustments of network anomaly detection algorithms. The experiment
can be deemed successful if validated at least with an open-source
implementation successfully applied with real networks.
3. Introduction
The main objective of a network anomaly detection system is to
identify Relevant States of the network as those are states that
could lead to problems or might be clear indications of problem
already happening. In order to formally define the statement above,
some definitions from [I-D.ietf-nmop-terminology] are reported in the
following:
Network Anomaly: An unusual or unexpected event or pattern in
network data in the forwarding plane, control plane, or management
plane that deviates from the normal, expected behavior.
Network problem: A state regarded as undesirable and which may
require remedial action.
State: A particular Condition that something (e.g., a Resource) has
(i.e., it is in a State) at a specific time.
Relevant State: A state that have relevancy for network operators,
as those are states that could lead to problems or might be clear
indications of problem already happening.
Riccobene, et al. Expires 4 September 2025 [Page 3]
Internet-Draft network-anomaly-lifecycle March 2025
Symptom: An observable Characteristic, State, Event, or Condition
considered as an indication of a Problem or potential Problem.
It is still remarkably difficult to gain a full understanding and a
complete perspective of "if" and "how" a relevant state is actually
an indication of a problem or it is just unexpected, but has no
impact on services and end users. Providers of solutions for network
anomaly detection should aim at increasing accuracy, by minimizing
false positives and false negatives. Moreover, the behaviour of the
network naturally changes over time, when more connectivity services
are deployed, more customers on-boarded, devices are upgraded or
replaced, and therefore it is almost impossible to identify anomaly
detection techniquest that can keep working accurately over time,
without changing the detection criterias (or methodologies) over
time.
This opens up to the necessity of further validating notified
relevant states to check if a detected symptom is actually impacting
connectivity services: this might require different actors (both
human and algorithmic) to act during the process and refine their
understanding across the network anomaly lifecycle.
Finally, once validation has happened, this might lead to refinements
to the logic that is used by the detection, so that this process can
improve the detection accuracy over time.
Performing network anomaly detection is a process that requires a
continuous learning and continuous improvement. Relevant states are
detected by aggregating and understanding Symptoms, then validated,
confirming that Symptoms actually impacted connectivity services
impacting and eventually need to be further analyzed by performing
postmortem analysis to identify any potential adjustment to improve
the detection capability. Each of these steps represents an
opportunity to learn and refine the process, and since
implementations of these steps might also be provided by different
parties and/or products, this document also contributes a formal data
model to capture and exchange Symptom information across the
lifecycle.
4. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
Riccobene, et al. Expires 4 September 2025 [Page 4]
Internet-Draft network-anomaly-lifecycle March 2025
This document makes use of the terms defined in
[I-D.ietf-nmop-terminology].
* State
* Problem
* Event
* Alarm
* Symptom
The following terms are used as defined in [RFC9417].
* Metric
* Intent
The following terms are defined in this document.
* Annotator: Is a human or an algorithm which produces metadata by
describing anomalies with Symptoms.
* False Positive: Is a detected anomaly which has been identified
during the postmortem to be not anomalous.
* False Negative: Is anomalous but has not been identified by the
anomaly detection system.
5. Defining Desired States
The above definitions of network problem provide the scope for what
to be looking for when detecting network anomalies. Concepts like
"desirable state" and "required state" are introduced. This poses
the attention on a significant problem that network operators have to
face: the definition of what is to be considered "desirable" or
"undesirable". It is not always easy to detect if a network is
operating in an undesired state at a given point in time. To
approach this, network operators can rely on different methodologies,
more or less deterministic and more or less sensitive: on the one
side, the definition of intents (including Service Level Objectives
and Service Level Agreements) which approaches the problem top-down;
on the other side, the definition of Symptoms, by mean of solutions
like SAIN [RFC9417], [RFC9418] and
[I-D.ietf-nmop-network-anomaly-architecture], which approaches the
problem bottom-up. At the center of these approaches, there are the
so-called Symptoms, explaining what is not working as expected in the
Riccobene, et al. Expires 4 September 2025 [Page 5]
Internet-Draft network-anomaly-lifecycle March 2025
network, sometimes also providing hints towards issues and their
causes.
One of the more deterministic approaches is to rely on Symptoms based
on measurable service-based KPIs, for example, by using Service Level
Indicators, Objectives and Agreements ([RFC9543]). This is the case
when rules on SLOs and SLIs are manually defined once and the used
afterwards for detection at runtime.
However, defining SLOs in a "static way" can bring some challenges as
well, related to the dynamic nature of networks and services.
Alternative methodologies rely on a more "relaxed" approach to detect
symptoms and their impact to services as a way to generate analytical
data out of operational data. For instance:
SAIN introduces the definition and exposure of Symptoms as a
mechanism for detecting those concerning behaviors in more
deterministic ways. Moreover, the concept of "impact score" has
been introduced by SAIN, to indicate what is the expected degree
of impact that a given Symptom will have on the services relying
on the related subservice to which the Symptom is attached.
Daisy introduces the concept of concern score to indicate what is
the degree of concern that a given Symptom could cause a
degradation for a connectivity service.
In general, defining boundaries between desirable vs. undesirable in
an accurate fashion requires continuous iterations and improvements
coming from all the stages of the network anomaly detection
lifecycle, by which network engineers can transfer what they learn
through the process into new Symptom definitions and, ultimately,
into refinements of the detection algorithms.
6. Lifecycle of a Network Anomaly
The lifecycle of a network anomaly can be articulated in three
phases, structured as a loop: Detection, Validation, Refinement.
Riccobene, et al. Expires 4 September 2025 [Page 6]
Internet-Draft network-anomaly-lifecycle March 2025
+-------------+
+--------> | Detection | ---------+
Adjustments | +-------------+ | Symptoms
| |
| v
+------------+ +------------+
| Refinement |<--------------------- | Validation |
+------------+ Problem +------------+
Confirmation
Figure 1: Anomaly Detection Refinement Lifecycle
Each of these phases can either be performed by a network expert or
an algorithm or complementing each other.
The network anomaly metadata is generated by an annotator, which can
be either a human expert or an algorithm. The annotator can produce
the metadata for a network anomaly, for each stage of the cycle and
even multiple versions for the same stage. In each version of the
network anomaly metadata, the annotator indicates the list of
Symptoms that are part of the network anomaly taken into account.
The iterative process is about the identification of the right set of
Symptoms.
6.1. Network Anomaly Detection
The Network Anomaly Detection stage is about the continuous
monitoring of the network through Network Telemetry [RFC9232] and the
identification of Symptoms. One of the main requirements that
operator have on network anomaly detection systems is the high
accuracy. This means having a small number of false negatives,
Symptoms causing connectivity service impact are not missed, and
false positives, Symptoms that are actually innocuous are not picked
up.
As the detection stage is becoming more and more automated for
production networks, the identified Symptoms might point towards
three potential kinds of behaviors:
i. those that are surely corresponding to an impact on connectivity
services, (e.g. the breach of an SLO),
ii. those that will cause problems in the future (e.g. rising trends
on a timeseries metric hitting towards saturation),
Riccobene, et al. Expires 4 September 2025 [Page 7]
Internet-Draft network-anomaly-lifecycle March 2025
iii. those or which the impact to connectivity services cannot be
confirmed (e.g. sudden increase/decrease of timeseries metrics,
anomalous amounts of log entries, etc.).
The first category requires immediate intervention (a.k.a. the
problem is "confirmed"), the second one provides pointers towards
early signs of an problem potentially happening in the near future
(a.k.a. the problem is "forecasted"), and the third one requires some
analysis to confirm if the detected Symptom requires any attention or
immediate intervention (a.k.a. the problem is "potential"). As part
of the iterative improvement required in this stage, one that is very
relevant is the gradual conversion of the third category into one of
the first two, which would make the network anomaly detection system
more deterministic. The main objective is to reduce uncertainty
around the raised alarms by refining the detection algorithms. This
can be achieved by either generating new Symptom definitions,
adjusting the weights of automated algorithms or other similar
approaches.
6.2. Network Anomaly Validation
The key objective for the validation stage is clearly to decide if
the detected Symptoms are signaling a real problem (a.k.a. requires
action) or if they are to be treated as false positives (a.k.a.
suppressing the alarm). For those Symptoms surely having impact on
connectivity services, 100% confidence on the fact that a network
problem is happening can be assumed. For the other two categories,
"forecasted" and "potential", further analysis and validation is
required.
6.3. Network Anomaly Refinement
After validation of a problem, the service provider performs
troubleshooting and resolution of the problem. Although the network
might be back in a desired state at this point, network operators can
perform detailed postmortem analysis of network problems with the
objective to identify useful adjustments to the prevention and
detection mechanisms (for instance improving or extending the
definition of SLIs and SLOs, refining concern/impact scores, etc.),
and improving the accuracy of the validation stage (e.g. automating
parts of the validation, implementing automated root cause analysis
and automation for remediation actions). In this stage of the
lifecycle it is assumed that the problem is under analysis.
After the adjustments are performed to the network anomaly detection
methods, the cycle starts again, by "replaying" the network anomaly
and checking if there is any measurable improvement in the ability to
detect problems by using the updated method.
Riccobene, et al. Expires 4 September 2025 [Page 8]
Internet-Draft network-anomaly-lifecycle March 2025
7. Introducing a Label Store for Network Anomaly labels
The information that is produced at each stage needs to be persisted
and retrieved to perform the network anomaly lifecycle. The
lifecycle begins with the detector notifying anomalies to the "Alarm
and Problem Management System" and to the "Post-mortem System"
according to (see [I-D.ietf-nmop-network-anomaly-architecture]). In
this case the Post-mortem system is identified as the Label Store.
Once the notification arrives to the Label Store, the anomaly label
is persisted. In the following stages (i.e. validation and
refinement), the information about the labels are retrieved, reviewd,
modified and persisted again, generating every time a new version of
the same annotation, or tagging the annotation as irrelevant, if it
would be necessary to remove it.
In the following sections, the following are defined: * a state
machine for a label * a YANG data model for the notification sent by
the Detector to the Label Store * a YANG data model to the define the
interrogation (and retrieval) of the labels from the label store.
8. Network Anomaly State Machine
In the context of this document, from a network anomaly detection
point of view a network problem is defined as a collection of
interrelated Symptoms, as specified in
[I-D.netana-nmop-network-anomaly-semantics].
The understanding of a network problem can change over time.
Moreover, multiple actors are involved in the process of refining
this understanding in the different phases.
From this perspective, a problem can be refined according to the
following states (Figure 2).
Riccobene, et al. Expires 4 September 2025 [Page 9]
Internet-Draft network-anomaly-lifecycle March 2025
+---------+
| Initial |-----------------+
+---------+ |
| |
+-----+---------+ |
+--------|---------------|------+ |
| +------v-----+ +------v----+ | |
| | Problem | | Problem | | |
+---->| | Forecasted | | Potential | | |
| | +------------+ +-----------+ | |
| +--------|--Detection---|-------+ |
| | | |
+-------+ | +------- ----- + |
| Final | | | |
+---^---+ | | |
| | | |
| | v |
| | +-----------Validation------------+ |
+-----------------------+ | | +-----------+ | |
| | | | | | Problem | | Problem | | |
| +-----------------+ | | | | Discarded | | Confirmed |<-|---+
| | Detection | | | | +-----|-----+ +-----------+ |
| | Adjusted |-------+ +---------------------------------+
| +--------^--------+ | | |
| | | | |
| | | +---v---+ |
| | | | Final | |
| | | +-------+ |
| +---------|--------+ | |
| | Problem | | |
| | Analyzed |<-|-----------------------------------+
| +------------------+ |
+-------Refinement------+
Figure 2: Network Anomaly State Machine
The knowledge gained at each stage is codified as a list of anomaly
labels that can be stored on a Label Store ( see Section 10.1 for a
reference).
9. Network Anomaly Data Model
9.1. Overview of the Data Model for the Relevant State and all the
related entities
Riccobene, et al. Expires 4 September 2025 [Page 10]
Internet-Draft network-anomaly-lifecycle March 2025
module: ietf-relevant-state
+--rw relevant-state
+--rw id yang:uuid
+--rw description? string
+--rw start-time yang:date-and-time
+--rw end-time? yang:date-and-time
+--rw concern-score score
+--rw anomalies* [id version]
+--rw id yang:uuid
+--rw uri? inet:uri
+--rw version yang:counter32
+--rw state identityref
+--rw description? string
+--rw start-time yang:date-and-time
+--rw end-time? yang:date-and-time
+--rw confidence-score score
+--rw pattern? identityref
+--rw annotator!
| +--rw name string
| +--rw (annotator-type)?
| +--:(human)
| | +--rw human? empty
| +--:(algorithm)
| +--rw algorithm? empty
+--rw symptom!
| +--rw id yang:uuid
| +--rw concern-score score
+--rw service!
+--rw id yang:uuid
notifications:
+---n relevant-state-notification
+--ro description? string
+--ro start-time yang:date-and-time
+--ro end-time? yang:date-and-time
+--ro concern-score score
+--ro anomalies* [id version]
+--ro id yang:uuid
+--ro uri? inet:uri
+--ro version yang:counter32
+--ro state identityref
+--ro description? string
+--ro start-time yang:date-and-time
+--ro end-time? yang:date-and-time
+--ro confidence-score score
+--ro pattern? identityref
+--ro annotator!
| +--ro name string
Riccobene, et al. Expires 4 September 2025 [Page 11]
Internet-Draft network-anomaly-lifecycle March 2025
| +--ro (annotator-type)?
| +--:(human)
| | +--ro human? empty
| +--:(algorithm)
| +--ro algorithm? empty
+--ro symptom!
| +--ro id yang:uuid
| +--ro concern-score score
+--ro service!
+--ro id yang:uuid
Figure 3: YANG tree diagram for ietf-relevant-state
<CODE BEGINS> file "ietf-relevant-state@2025-03-03.yang"
module ietf-relevant-state {
yang-version 1.1;
namespace "urn:ietf:params:xml:ns:yang:ietf-relevant-state";
prefix rsn;
import ietf-yang-types {
prefix yang;
reference "RFC 6021: Common YANG Data Types";
}
import ietf-inet-types {
prefix inet;
reference "RFC 6991: Common YANG Data Types";
}
organization
"IETF NMOP Working Group";
contact
"WG Web: <https://datatracker.ietf.org/wg/nmop/>
WG List: <mailto:nmop@ietf.org>
Authors: Vincenzo Riccobene
<mailto:vincenzo.riccobene@huawei-partners.com>
Thomas Graf
<mailto:thomas.graf@swisscom.com>
Wanting Du
<mailto:wanting.du@swisscom.com>
Alex Huang Feng
<mailto:alex.huang-feng@insa-lyon.fr>";
description
"This module defines the relevant-state container and
notifications to be used by a network anomaly detection
system. The defined objects can be used to augment
Riccobene, et al. Expires 4 September 2025 [Page 12]
Internet-Draft network-anomaly-lifecycle March 2025
operational network collected observability data and
analytical problem data equally. Describing the relevant-state
of observed symptoms.
Copyright (c) 2025 IETF Trust and the persons identified as
authors of the code. All rights reserved.
Redistribution and use in source and binary forms, with or
without modification, is permitted pursuant to, and subject
to the license terms contained in, the Revised BSD License
set forth in Section 4.c of the IETF Trust's Legal Provisions
Relating to IETF Documents
(https://trustee.ietf.org/license-info).
This version of this YANG module is part of RFC XXXX; see the RFC
itself for full legal notices.";
revision 2025-03-03 {
description
"Initial version";
reference
"RFC XXX: Semantic Metadata Annotation for Network Anomaly Detection";
}
typedef score {
type uint8 {
range "0 .. 100";
}
description "Number that indicates a score between 0 and 100";
}
identity network-anomaly-state {
description
"Base identity for representing the state of the network anomaly";
}
identity detection {
base network-anomaly-state;
description
"A problem reached detection state";
}
identity validation {
base network-anomaly-state;
description
"A problem reached validation state";
}
identity refinement {
base network-anomaly-state;
description
"A problem reached refinement state";
Riccobene, et al. Expires 4 September 2025 [Page 13]
Internet-Draft network-anomaly-lifecycle March 2025
}
identity problem-forecasted {
base detection;
description
"A problem has been forecasted, as it is expected that
the indicated list of symptoms will impact a service
in the near future";
}
identity problem-potential {
base detection;
description
"A problem has been detected with a confidence
lower than 100%. In order to confirm that this set of
symptoms are generating service impact, it requires further
validation";
}
identity problem-confirmed {
base validation;
description
"After validation, the problem has been confirmed";
}
identity discarded {
base validation;
description
"After validation, the network anomaly has been
discarded, as there is no evindence that it is causing an
problem";
}
identity analyzed {
base refinement;
description
"The anomaly detection went through analysis to identify
potential ways to further improve the detection process in
for future anomalies";
}
identity adjusted {
base refinement;
description
"The network anomaly has been solved and analysed.
No further action is required.";
}
identity pattern {
description
"Pattern identified by the Detector.";
}
identity drop {
description
Riccobene, et al. Expires 4 September 2025 [Page 14]
Internet-Draft network-anomaly-lifecycle March 2025
"Drop of the value";
}
identity spike {
description
"Spike of the value";
}
identity mean-shift {
description
"Shift of the mean of the value";
}
identity seasonality-shift {
description
"Shift of the seasonality of the value";
}
identity trend {
description
"Trend exhibited by the value";
}
identity other {
description
"Any other type of pattern";
}
grouping relevant-state-grouping {
description "Relevant State is a state that could lead to
problems or might be clear indications of problem already happening";
leaf description {
type string;
description
"Textual description of the fault";
}
leaf start-time {
type yang:date-and-time;
mandatory true;
description
"Date and time indicating the beginning of the fault";
}
leaf end-time {
type yang:date-and-time;
description
"Date and time indicating the end of the fault";
}
leaf concern-score {
type score;
mandatory true;
description
"Score indicating the degree of concern in
relation to the overall relevant state.";
Riccobene, et al. Expires 4 September 2025 [Page 15]
Internet-Draft network-anomaly-lifecycle March 2025
}
}
grouping annotator-grouping {
description "Annotator represents the entity that produced the
annotation (it is either a human or an algorithm)";
leaf name {
type string;
mandatory true;
description
"Name of the annotator (either user or algorithm)
If it is an algorithm, the name can also include
the version.";
}
choice annotator-type {
description "An annotator can be either a human user or a
programmatic entity, such as an algorithm";
case human {
leaf human {
type empty;
description
"This option is used if a human provided the label";
}
}
case algorithm {
leaf algorithm {
type empty;
description
"This option is used if a software provided the label";
}
}
}
}
grouping anomaly-grouping {
description "List of anomalies that are part of the relevant state";
list anomalies {
key "id version";
description "List of Anomaly instances";
leaf id {
type yang:uuid;
description
"Unique ID of the anomaly";
}
leaf uri {
type inet:uri;
description
Riccobene, et al. Expires 4 September 2025 [Page 16]
Internet-Draft network-anomaly-lifecycle March 2025
"URI to viusalize the analytical metrics of the anomaly.";
}
leaf version {
type yang:counter32;
description
"Version of the anomaly metadata object.
It allows multiple versions of the metadata to be
generated in order to support the definition of
multiple problem objects from the same source to
facilitate improvements overtime";
}
leaf state {
type identityref {
base network-anomaly-state;
}
mandatory true;
description "State of the anomaly";
}
leaf description {
type string;
description
"Textual description of the anomaly";
}
leaf start-time {
type yang:date-and-time;
mandatory true;
description
"Date and time indicating the beginning of the anomaly
A detection system will alwasys set a start time,
as it represents the moment in time from which the
behaviour of the monitored system is considered
to be anomalous with respect its expected behaviour";
}
leaf end-time {
type yang:date-and-time;
description
"Date and time indicating the end of the anomaly.
This field is indicated as non mandatory, as it could
be the case that the anomaly is still happening at the
time of generation of the label";
}
leaf confidence-score {
type score;
mandatory true;
description "Score indicating how confident was the detector
while considering the given anomaly as part of the relevant
event";
}
Riccobene, et al. Expires 4 September 2025 [Page 17]
Internet-Draft network-anomaly-lifecycle March 2025
leaf pattern {
type identityref {
base pattern;
}
description
"Pattern describes the type of pattern that was
detected by the annotator (e.g. spike, drop, mean-shift,
etc.)";
}
container annotator {
presence "It specifies an annotator for the anomaly";
description
"Annotator represents the entity that produced the
annotation";
uses annotator-grouping;
}
container symptom {
presence "It specifies the symptom for the anomaly";
description
"An observable Characteristic, State, Event, or Condition
considered as an indication of a Problem or potential
Problem";
leaf id {
type yang:uuid;
mandatory true;
description
"Unique ID of the symptom type";
}
leaf concern-score {
type score;
mandatory true;
description
"Score indicating the degree of concern in
relation to the specific symptom. Each
symptom will carry a certain degree of
concern that is specific to the symptom.";
}
}
container service {
presence
"It specifies the service (or the monitored entity)
affected by the anomaly";
description
"Service represents the entity that is monitored, which
for instance can be a connectivity servise, such as an
L3VPN";
Riccobene, et al. Expires 4 September 2025 [Page 18]
Internet-Draft network-anomaly-lifecycle March 2025
leaf id {
type yang:uuid;
mandatory true;
description
"Unique ID of the service (or monitored entity)
This is supposed to be augmented by other modules
that want to define the service affected by the
anomaly";
}
}
}
}
notification relevant-state-notification {
description
"Notification of a relevant state that can be sent by the
anomaly detection system to the postmortem management system or to the
incident management system";
uses relevant-state-grouping;
uses anomaly-grouping;
}
container relevant-state {
description "A Relevant State is a state that have relevancy
for network operators, as those are states that could lead
to problems or might be clear indications of problem already
happening";
leaf id {
type yang:uuid;
mandatory true;
description
"Unique ID of the relevant state
It is unique in the scope of the Label Store";
}
uses relevant-state-grouping;
uses anomaly-grouping;
}
}
<CODE ENDS>
Figure 4: YANG module for ietf-relevant-state
The data model provides support for "human-in-the-loop", allowing for
network experts to validate and adjust network anomaly labels and
detection systems. An example of human-in-the-loop has been
demonstrated with Antagonist [Antagonist], by building a User
Interface that interacts with an API based on this data model.
Riccobene, et al. Expires 4 September 2025 [Page 19]
Internet-Draft network-anomaly-lifecycle March 2025
The base for the modules is the relevant-state data model. Relevant
state is at the root of the data model, with its parameters (ID,
description, start-time, end-time) and a collection of anomalies.
This allows the relevant state to be considered as a container of
anomalies.
Each anomaly is characterized by some intrinsic fields (such as id,
version, state, description, start-time, end-time, confidence score
and pattern) Particularly the confidence score is a measure of how
confident was the detector in considering the given anomaly as an
anomalous behaviour.
Each anomaly also include the symptom and the service container.
These containers are placeholders to represent the information about
the symptom (what is exaclty happening as anomalous behaviour) and
the connectivity service (what entity is affected by the anomaly).
In particular, for what concerns the symptom, a concern score is
defined as necessary field, which has the meaning of expressing how
much the anomaly is impacting connectivity services. In case
additional information related to the symptom and to the service need
to be provided, augmentation would be the appropriate intended
mechanism to do so. An example of this is provided in
[I-D.netana-nmop-network-anomaly-semantics], where an augmentation of
both symptom and service is provided for the specific case of anoamly
labels related to connectivity services.
Also a list of various actors that can be involved in the process is
presented as following:
In the detection stage: the detectors can be Network Engineers and/
or Automatic detectors (including Rule-based detectors and ML-
based detectors)
In the validation stage: the validators can be Network Engineers
manually validating the labels
In the refinement stage: the refiners can be Data Scientists and/or
Automatic Refiners (including systems that automatically refine
the detection systems, based on the validated labels).
The data model that has been defined is used in two YANG modules: the
relevant-state-notification and the ietf-relevant-state: the
notification is primarily used by the Network Anomaly Detector, to
notify the "Alarm and Problem Management System" and the "Post-mortem
System" (see [I-D.ietf-nmop-network-anomaly-architecture]); the
container instead is used inside the Post-mortem system to exhance
anomaly detection lables between the anomaly detection stages defined
above (validation, refinement, detection).
Riccobene, et al. Expires 4 September 2025 [Page 20]
Internet-Draft network-anomaly-lifecycle March 2025
10. Implementation status
This section provides pointers to existing open source
implementations of this draft. Note to the RFC-editor: Please remove
this before publishing.
10.1. Antagonist
An open-source implementation for this draft is called AnTagOnIst
(Anomaly Tagging On hIstorical data), and it has been implemented in
order to validate the application of the YANG model defined in this
draft. Antagonist provides visual support for two important use
cases in the scope of this document:
* the generation of a ground truth in relation to symptoms and
problems in timeseries data
* the visual validation of results produced by automated network
anomaly detection tools.
The open-source code can be found here: [Antagonist]
As part of the experiment that was conducted with AnTagOnIst, Some
main Use Case scenarios have been validated so far:
Exposure of a GUI for human validation of the labels.
Integration with Rule Based anomaly detection systems. In
particular the integration with SAIN and Cosmos Bright Lights is
ongoing.
Integration with ML-based detection systems.
11. Security Considerations
The security considerations will have to be updated according to
"https://wiki.ietf.org/group/ops/yang-security-guidelines".
12. Acknowledgements
The authors would like to thank Antonio Roberto for his contribution
to the ideas in this draft and Mohamed Boucadair for his review and
valuable comments.
13. Normative References
Riccobene, et al. Expires 4 September 2025 [Page 21]
Internet-Draft network-anomaly-lifecycle March 2025
[Antagonist]
Riccobene, V., Du, W., Graf, T., and H. Huang Feng,
"Antagonist: Anomaly tagging on historical data",
<https://github.com/vriccobene/antagonist>.
[I-D.ietf-nmop-network-anomaly-architecture]
Graf, T., Du, W., and P. Francois, "An Architecture for a
Network Anomaly Detection Framework", Work in Progress,
Internet-Draft, draft-ietf-nmop-network-anomaly-
architecture-01, 20 October 2024,
<https://datatracker.ietf.org/doc/html/draft-ietf-nmop-
network-anomaly-architecture-01>.
[I-D.ietf-nmop-terminology]
Davis, N., Farrel, A., Graf, T., Wu, Q., and C. Yu, "Some
Key Terms for Network Fault and Problem Management", Work
in Progress, Internet-Draft, draft-ietf-nmop-terminology-
12, 22 February 2025,
<https://datatracker.ietf.org/doc/html/draft-ietf-nmop-
terminology-12>.
[I-D.netana-nmop-network-anomaly-semantics]
Graf, T., Du, W., Feng, A. H., Riccobene, V., and A.
Roberto, "Semantic Metadata Annotation for Network Anomaly
Detection", Work in Progress, Internet-Draft, draft-
netana-nmop-network-anomaly-semantics-04, 3 November 2024,
<https://datatracker.ietf.org/doc/html/draft-netana-nmop-
network-anomaly-semantics-04>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[RFC8340] Bjorklund, M. and L. Berger, Ed., "YANG Tree Diagrams",
BCP 215, RFC 8340, DOI 10.17487/RFC8340, March 2018,
<https://www.rfc-editor.org/info/rfc8340>.
[RFC9232] Song, H., Qin, F., Martinez-Julia, P., Ciavaglia, L., and
A. Wang, "Network Telemetry Framework", RFC 9232,
DOI 10.17487/RFC9232, May 2022,
<https://www.rfc-editor.org/info/rfc9232>.
Riccobene, et al. Expires 4 September 2025 [Page 22]
Internet-Draft network-anomaly-lifecycle March 2025
[RFC9417] Claise, B., Quilbeuf, J., Lopez, D., Voyer, D., and T.
Arumugam, "Service Assurance for Intent-Based Networking
Architecture", RFC 9417, DOI 10.17487/RFC9417, July 2023,
<https://www.rfc-editor.org/info/rfc9417>.
[RFC9418] Claise, B., Quilbeuf, J., Lucente, P., Fasano, P., and T.
Arumugam, "A YANG Data Model for Service Assurance",
RFC 9418, DOI 10.17487/RFC9418, July 2023,
<https://www.rfc-editor.org/info/rfc9418>.
[RFC9543] Farrel, A., Ed., Drake, J., Ed., Rokui, R., Homma, S.,
Makhijani, K., Contreras, L., and J. Tantsura, "A
Framework for Network Slices in Networks Built from IETF
Technologies", RFC 9543, DOI 10.17487/RFC9543, March 2024,
<https://www.rfc-editor.org/info/rfc9543>.
Authors' Addresses
Vincenzo Riccobene
Huawei
Dublin
Ireland
Email: vincenzo.riccobene@huawei-partners.com
Thomas Graf
Swisscom
Binzring 17
CH-8045 Zurich
Switzerland
Email: thomas.graf@swisscom.com
Wanting Du
Swisscom
Binzring 17
CH-8045 Zurich
Switzerland
Email: wanting.du@swisscom.com
Alex Huang Feng
INSA-Lyon
Lyon
France
Email: alex.huang-feng@insa-lyon.fr
Riccobene, et al. Expires 4 September 2025 [Page 23]