A glossary of industry and Augtera Networks terms including: Network AI, AIOps, Network AIOps, Autocorrelation.
Augtera Network AI / platform specific terms will be labelled (Augtera).
The application of AI/ML to an area of IT Operation automation, characterized by big data, anomaly detection, correlation, and causality determination.
AIOPs for networking
Specialized AIOps for Networking. See AIOps.
AI Needles (Augtera)
In Network AI, AI needles are the result of automating the mining of the data haystack to produce proactive, operationally relevant, actionable insights.
Exhibiting human-like intuition, judgement, or capabilities. Exhibiting human-like intuition, judgement, or capabilities. Augtera’s Network AI can group events / anomalies into a single incident, identify incident root, and understand syslog semantics through natural language processing. Augtera’s Network AI includes 9+ proprietary machine learning algorithms.
Assisted Correlation (Augtera)
Customer-driven correlation through the Augtera platform UI/UX. See multi-layer assisted correlation.
Automated topology-aware correlation. See multi-layer autocorrelation.
Automated Anomaly Detection (Augtera)
See Autonomous Anomaly Detection
Automated Log Analysis (Augtera)
Augtera automatically classifies log data and quickly detects pattern changes and behavioral anomalies that other systems can’t find. Augtera also leverages proprietary NLP and Machine Learning to find zero-day anomalies in log data. These anomalies enable automatic detection and notification of operationally relevant logs when they first occur
Autonomous Anomaly Detection (Augtera)
In Augtera’s Network AI, 9+ proprietary, network-specific, ML algorithms built from the ground up that autonomously learn the patterns of all network infrastructure, application flow and SLA timeseries metrics as well as syslog text, build online models and predict operational anomalies in real-time. This enables proactive detection before brewing network issues and grey failures explode into downtime and business impact.
In statistics, correlation is the assessment of the relationship between two or more variables. Variables can be positively or negatively correlated. Variables can also have no correlation.
In Network AI, autocorrelation is the automated assessment of relationship between two or more events / anomalies as a way of grouping them into a single incident to:
- Reduce notification noise.
- Determine, with the use of the network model, the incident root and consequences.
In Network AI, assisted correlation is customer-driven correlation through the Augtera UI / UX.
Data Haystack (Augtera)
The pile of unsorted and often unseen data the occurs as collecting traditional and new forms of streaming data. Billions of data points a day can be collected, but without systematic and real-time analysis, they become noise points. Obfuscating what is operationally relevant and rarely looked at, except in ad-hoc forensics searches, for already occurred failures.
Augtera’s Network AI transforms the manual, reactive, and noisy approach to human mining of the haystack into a process that is automated, proactive, and operationally relevant.
Augtera tends to use the term “incident” to refer to multiple related events and anomalies. To reduce noise, Augtera creates parent trouble tickets and operations alerts at the incident level.
Viewed as an aspect of Artificial Intelligence, machine learning refers to algorithms that improve through the use of data. The Augtera platform uses machine learning to create models of network-specific patterns, make predictions based on those patterns, and to identify variations from those patterns.
Multi-Layer Assisted Correlation (Augtera)
Physical and virtual topology visualizations with an overlay of network events and anomalies across multiple data sources with time machine support. This assists the operator to rapidly diagnose the root cause when there are complaints from application
Multi-Layer AutoCorrelation (Augtera)
In Augtera’s Network AI, multi-layer correlation is a proprietary machine learning algorithm that is multi-layer network topology aware and correlates network events and Augtera anomalies across multiple data sources automatically. This enables operators to proactively receive notifications of correlated network issues with high fidelity context, further reducing ticketing noise and enabling rapid root cause analysis and remediation.
Network AI (Augtera)
AI/ML specialized for networking. Network / network operations data is different in nature from data studied & trained on by off-the-shelf / academic AI/ML. The distributions are different from network to network. Augtera has developed 9+ proprietary algorithms to ensure that noiseless anomaly detection, correlation, and notification can be performed, resulting in high-fidelity, operationally relevant, actionable insights.
AIOps for networking. See AIOps.
Network Model (Augtera)
In Augtera’s Network AI, the Network Model is the holistic modeling of multilayer physical and logical network objects. All nodes, relationships and attributes. The comprehensive approach is distinct from modeling just the routing topology or associating network objects on a graph based on how they appear in error messages.
Augtera’s approach enables Network AI to understand hierarchy, for example how a link interface issue impacts a BGP session failure, as well as topology-aware correlation.
Networking or computer networking refers to the transfer of information between endpoints, across connected devices such as routers and switches. Endpoints are varied but include smart phones, laptops, and applications.
Real-Time Syslog (Augtera)
Augtera’s solution for analyzing billions of streaming syslog messages per day, detecting anomalies, and/or taking an action based on classifiers. The solution includes real-time natural language processing to detect uniquely new messages.
In Augtera’s Network AI, spaces are a way for customers to define / restrict what correlations are performed. This allows customers to specify what is most operationally relevant to them.
Supervised and Unsupervised Learning
The common approaches to machine learning are supervised and unsupervised learning. Supervised learning creates patterns from data that is “labelled” (this is a cat, this is a dog,…). Unsupervised learning detects patterns in data that is not labelled.
Network use cases most often call for unsupervised learning because a) labelled data is hard to come by and b) more importantly, traffic patterns are different on every interface and every network. It is difficult to apply the patterns from one network to another.
While some consider unsupervised learning not as accurate as supervised learning, Augtera Networks has invested 4+ years of R&D and real-world experience to develop 9+ proprietary algorithms specialized for network use cases. These algorithms are part of an overall approach to noise reduction.
A threshold implies a single finite measurement / target level. For example, in many systems, a single temperature threshold might be set, and an alarm generated if the temperature goes above that threshold. Augtera’s Network AI does not use thresholds because they generate false positives or false negatives. Instead, machine learning models are used that understand network patterns / distributions.
See Supervised and Unsupervised Learning.
In Augtera’s Network AI, views are a way for customers to define / restrict what is displayed in the UI / UX and / or notified to a console such as Slack or an operations tool such as ServiceNow. Views are powerful ways to explore data and investigate problems. The same mechanism is a powerful way of defining what is operational relevant to be notified on.
Zero Day Anomalies (Augtera)
Syslog messages being detected for the first time, and therefore potential anomalies. After inspection, if network operations determines the new message is of interest, a classifier can be quickly created and installed without requiring a software update.