• Ingestion of metrics, events, logs, and traces
• Faults, events, and time-series data enriched with inventory, topology, and service dependencies
• Monitoring within and across applications, and service domains
• Automatic generation of baselines using unsupervised machine learning
• Stochastic modeling that enables more effective noise reduction
• Affinity analysis to group signals within and across service layers
• Ontology discovery that provides richer meta data for root cause analysis
• Correlating issues to the customers’ experience and prioritizing actions based on impact

Fault and Performance Management in a single application

Managing service assurance independently within each service layer constrains operational productivity and the service assurance process. Teams are often chasing symptoms within their silo when the root cause of the problem lies outside their visibility and control. Multiple operations support teams may be addressing the same or different symptoms but all of them attributable to the same root cause.  Addressing these service issues is slow and extremely labor intensive.

VIA AIOps delivers the information required to manage faults and performance across the service delivery stack in one application with a single incident inbox.

VIA AIOps automatically determines the correct algorithm to use on streaming data to generate baselines and better detect anomalies.  Unsupervised machine learning generates baselines for every metric and dimension combination.

Baselines are created as data is ingested and then continuously updated and improved over time with more data. The use of machine-learned baselines versus simple thresholds allows VIA to reduce noise and detect anomalous behavior occurring during both low usages and peak periods.

These baselines reduce noise efficiently and identify anomalies faster.

Anomalous signals from data streams being monitored are often temporary usage spikes or statistical noise. The ability to identify anomalies that are significant and non-transient enables operations teams to focus on those problems that truly need fixes.

Stochastic models excel at separating signal from noise. These models can continuously monitor and evaluate suspicious changes in behavior of every metric, event, and entity. Stochastic models correctly detect the patterns those other techniques typically misclassify, identify late, or miss altogether.  Here are some examples of behavioural changes that are caught early using our modelling techniques.

With only a handful of out of the box algorithms, you quickly gain the advantages of VIA’s deduplication, alarm noise reduction, and incident prioritization for all hosts, entities, or services in your operating environment.  Artificial Intelligence and machine learning embedded within VIA’s out of the box algorithms are explainable, extensible, and easy to use.

On day one, each client can improve fault management performance with no configuration required by the end user and achieve 99% alarm noise reduction.

The signal onboarding process builds data models automatically with dimensions and data types.  It creates KPIs and metrics from the data as needed such as continuous counters, applying math functions or by deriving new metrics from the combination of multiple data streams.

This simplifies and speeds the processing of data preparation, ingestion, and enrichment. It eliminates the manual process steps required to cleanse, structure, and develop code for data construction.  It allows data to be ingested without having to fit a specific data model or data specification.

VIA’s feedback loop combines human and artificial intelligence to continuously improve accuracy of AI models and support the enablement of automated actions.  VIA provides information on each incident it discovers including the probable root cause, key symptoms, impacted populations, duration, and severity.  Users can provide feedback and provide context.  The system uses this information for relevance ranking and training the algorithm for greater accuracy.  In addition, this information over time can be used to determine if automated action should be implemented based on the consistency of the response needed to respond to the incident.

To further maximize operational productivity, dashboards can be created for all users from executives who need a high level view of service performance to the service reliability engineers who need to take actions on the service performance issues identified.