Semantic Metadata Annotation for Network Anomaly Detection

Network Anomaly Detection Architecture provides an overall introduction into how anomaly detection is being applied into the IP network domain and which operational data is needed. It approaches the problem space by automating what a Network Engineer would normally do when verifying a network connectivity service. Monitor from different network plane perspectives to understand wherever one network plane affects another negatively. In order to fine tune outlier detection, the results provided as analytical data need to be reviewed by a Network Engineer. Keeping the human out of the monitoring but still involving him in the alert verification loop. This document describes what information is needed to understand the output of the outlier detection for a Network Engineer, but also at the same time is semantically structured that it can be used for outlier detection testing by comparing the results systematically and set a baseline for supervised machine learning which requires labeled operational data.

Outlier Detection, also known as anomaly detection, describes a systematic approach to identify rare data points deviating significantly from the majority. Outliers can manifest as single data point or as a sequence of data points. There are multiple ways in general to classify anomalies, but for the context of this draft, the following three classes are taken into account:

Global outliers:: An outlier is considered "global" if its behaviour is outside the entirety of the considered data set. For example, if the average dropped packet count is between 0 and 10 per minute and a small time-window gets the value 1000, this is considered a global anomaly.

Contextual outliers:: An outlier is considered "contextual" if its behaviour is within a normal (expected) range, but it would not be expected based on some context. Context can be defined as a function of multiple parameters, such as time, location, etc. For example, the forwarded packet volume overnight reaches levels which might be totally normal for the daytime, but anomalous and unexpected for the nighttime.

Collective outliers:: An outlier is considered "collective" if the behaviour of each single data point that are part of the anomaly are within expected ranges (so they are not anomalous it either a contextual or a global sense), but the group, taking all the data points together, is. Note that the group can be made within a single time series (a sequence of data points is anomalous) or across multiple metrics (e.g. if looking at two metrics together, the combined behavior turns out to be anomalous). In Network Telemetry time series, one way this can manifest is that the amount of network paths and interface state changes matches the time range when the forwarded packet volume decreases as a group.

For each outlier a score between 0 and 1 is being calculated. The higher the value, the higher the probability that the observed data point is an outlier. Anomaly detection: A survey gives additional details on anomaly detection and its types.

The Data Mesh Architecture distinguishes between operational and analytical data. Operational data refers to collected data from operational systems. While analytical data refers to insights gained from operational data. In terms of network observability, semantics of operational network metrics are defined by IETF and are categorized as described in the Network Telemetry Framework in the following three different network planes:

Management Plane:: Time series data describing the state changes and statistics of a network node and its components. For example, Interface state and statistics modelled in ietf-interfaces.yang

Control Plane:: Time series data describing the state and state changes of network reachability. For example, BGP VPNv6 unicast updates and withdrawals exported in BGP Monitoring Protocol (BMP) and modeled in BGP

Forwarding Plane:: Time series data describing the forwarding behavior of packets and its data-plane context. For example, dropped packet count modelled in IPFIX entity forwardingStatus(IE89) and packetDeltaCount(IE2) and exportet with IPFIX .

In terms of network observability, this applies to operational semantic metadata and service level indicators. The health status and symptoms described in the Service Assurance Intend Based Networking , the precision availability metrics defined in or network anomalies and its symptoms as described in this document and applied in the network anomaly postmortem lifecycle described in where the applied semantic metadata of this document is refined for each detected anomaly.

In this section observed network symptoms are specified and categorized according to the following scheme:

Action:: Which action the network node performed for a packet in the Forwarding Plane, a path or adjacency in the Control Plane or state or statistical changes in the Management Plane. For Forwarding Plane we distinguish between missing, where the drop occured outside the measured network node, drop and on-path delay, which was measured on the network node. For Control Plane we distinguish between reachability, which refers to a change in the routing or forwarding information base (RIB/FIB) and adjcacency which refers to a change in peering or link-layer resolution. For Management Plane we refer to state or statistical changes on interfaces.

Reason:: For each action, one or more reasons describe why this action was used. For Drops in Forwarding Plane we distinguish between Unreachable because network layer reachability information was missing, Administered because an administrator configured a rule preventing the forwarding for this packet and Corrupt where the network node was unable to determine where to forward to due to packet, software or hardware error. For on-path delay we distinguish between Minimum, Average and Maximum Delay for a given flow. For Control Plane wherever a the reachability was updated or withdrawn or the adjcacency was established or teared down. For Management Plane we distinguish between interfaces states up and down, and statistical erros, discards or unknown protocol counters.

Cause:: For each reason one or more cause describe the cause why the network node has chosen that action.

consolidates for the forwarding plane a list of common symptoms with their Actions, Reasons and Causes. Describing Symptoms and their Actions, Reason and Cause for Forwarding Plane

Action	Reason	Cause
Missing	Previous	Time
Drop	Unreachable	next-hop
Drop	Unreachable	link-layer
Drop	Unreachable	Time To Life expired
Drop	Unreachable	Fragmentation needed and Don't Fragment set
Drop	Administered	Access-List
Drop	Administered	Unicast Reverse Path Forwarding
Drop	Administered	Discard Route
Drop	Administered	Policed
Drop	Administered	Shaped
Drop	Corrupt	Bad Packet
Drop	Corrupt	Bad Egress Interface
Delay	Min	-
Delay	Mean	-
Delay	Max	-

consolidates for the control plane a list of common symptoms with their actions, reasons and causess. Describing Symptoms and their Actions, Reason and Cause for Control Plane

Action	Reason	Cause
Reachability	Update	Imported
Reachability	Update	Received
Reachability	Withdraw	Received
Reachability	Withdraw	Peer Down
Reachability	Withdraw	Suppressed
Reachability	Withdraw	Stale
Reachability	Withdraw	Route Policy Filtered
Reachability	Withdraw	Maximum Number of Prefixes Reached
Adjacency	Established	Peer
Adjacency	Established	Link-Layer
Adjacency	Locally Teared Down	Peer
Adjacency	Remotely Teared Down	Peer
Adjacency	Locally Teared Down	Link-Layer
Adjacency	Remotely Teared Down	Link-Layer
Adjacency	Locally Teared Down	Administrative
Adjacency	Remotely Teared Down	Administrative
Adjacency	Locally Teared Down	Maximum Number of Prefixes Reached
Adjacency	Remotely Teared Down	Maximum Number of Prefixes Reached
Adjacency	Locally Teared Down	Transport Connection Failed
Adjacency	Remotely Teared Down	Transport Connection Failed

consolidates for the management plane a list of common symptoms with their Actions, Reasons and Causes. Describing Symptoms and their Actions, Reason and Cause for Management Plane

Action	Reason	Cause
Interface	Up	Link-Layer
Interface	Down	Link-Layer
Interface	Errors	-
Interface	Discards	-
Interface	Unknown Protocol	-

Metadata adds additional context to data. For instance, in networks the software version of a network node where Management Plane metrics are obtained from as described in. Where in Semantic Metadata the meaning or ontology of the annotated data is being described. In this section a YANG model is defined in order to provide a structure for the metadata related to anomalies happening in the network. The module is intended to describe the metadata used to "annotate" the operational data collected from the network nodes, which can include time series data and logs, as well as other forms of data that is "time-bounded". The aspects discussed so far in this document are grouped under the concept of "anomaly" which represents a collection of symptoms. The anomaly overall has a set of parameters that describe the overall behavior of the network in a given time-window including all the spotted symptoms (network anomalies).

contains the YANG tree diagram of the ietf-symptom-semantic-metadata module. For each symptom, the following parameters have been assigned: A unique ID for identification, a description of the symptom, a list of affected metrics or counters, atart and end time to specify the time-window, a confident score indicating how accurate the symptom was detected, a concern score indicating how critical the symptom is, the source indicating if it has been identified by a network expert or an algorithm, the tags with key value where Action, Reason and Cause can be annotated as described in previous section.

The security considerations.

This section provides pointers to existing open source implementations of this draft. Note to the RFC-editor: Please remove this before publishing.

A tool called Antagonist has been implemented during the IETF 119 Hackathon, in order to validate the application of the YANG models 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 incidents in timeseries data
the visual validation of results produced by automated network anomaly detection tools.

The open source code can be found here:

The authors would like to thank xxx for their review and valuable comments.