Introduction
According to a recent survey done by Enterprise Management Associates, 81% of Network Practitioners do not associate AIOps / Machine Learning with Observability.
AIOps that is purpose-built for Networking, is a generational step forward, going well beyond monitoring and observability to achieve new outcomes: dramatic reduction in alert / trouble ticket fatigue, multi-layer incident root identification, cloud-native geo-diverse scaling & HA, automation playbooks, and more.
Thresholds vs Algorithms
There are some metrics and signatures for which thresholds / rules work well. These are static definitions that are always “true”. For example, a Network Operations team may have a policy that it wants to be alerted whenever packet loss goes above 1%. On the other hand, setting and maintaining latency thresholds for links with varying distances and loads, is time consuming, prone to error, and often produces many false positives, or false negatives, depending on how they are set. AIOps machine learning algorithms learn patterns and report meaningful / relevant deviations from patterns that humans have difficulty spotting. Learn more about thresholds and ML algorithms in our previous blog: Machine Learning Anomaly Detection – Beyond Thresholds.
Reducing Alert and Trouble Ticket Fatigue
It’s no secret that one of the biggest problems plaguing network operations teams is the number of generated alerts / trouble tickets from an average of 4-10 networking tools per team. Network AIOps is a game changer. One source of false positive reduced truth, across all data sources, of what is an anomaly, what is a redundant alert / trouble ticket, and what is the lowest layer incident root object. ML algorithms that produce less false positives. Multi-layer autocorrelation that reduces redundant alarms / tickets, and better models’ relationships across all layers, to identify incident root. Learn more about Noise elimination in our previous blog: Noise Elimination in Network AI and the case study: Fortune 500 Data Center Solution.
Flexible Deployment
Software platforms / services built in the last-few years are fundamentally different than those built a decade or more ago, and/or cobbled together from tens of acquisitions. Today’s leading-edge Network AIOps platforms have cloud-native and cloud-independent architectures enabling horizontal scaling, of geo-diverse active-active high availability architectures, as SaaS, Hybrid Cloud, or On-prem deployments. SaaS reduces many startup and management costs for teams that value those characteristics, while on-prem meets the needs of those companies that are not comfortable with their data being in the “cloud”. Flexible deployment, geo-diverse high availability, and scalability are just some of the attributes of next generation software architectures.
Automation Playbooks
Automation is the end goal of Network Operations today. Agility, productivity, and error-reduction. However, every team is taking this journey at their own pace. Leading Network AIOps tools will have a “plug-in” architecture that allows customers to consume events, alerts, and other operations data, so their own automation playbooks can decide next steps.
Real-Time Logs
There are many log solutions that retain every log message, for long periods of time, with the ability to query archives. Great for compliance, deep-dive analysis of difficult to understand issues, and more. However, this is not leveraging logs for real-time anomaly detection, metric extraction, burst/rate-change detection, rare message detection, signature matching, and multi-layer, multi-data type auto correlation, at the speed of streaming data. That requires a totally different architecture from a logic, performance, and efficiency perspective. Learn more about how the Augtera Network AI approach to log analysis is different than existing log solutions in our previous blog: LogAI vs Existing Log Solutions.
Network AIOps is different than IT AIOps
Network AIOps has some overlap with IT AIOps, however it is significantly different, purpose-built for the Network Operations mission.
To give some examples of a difference this makes to auto-discover the physical, optical, Ethernet, and IP topologies, it is important to have a robust SNMP collection capability. There is often no equivalent dynamic REST API capability. Similarly, IT AIOps solutions do not support network-specific interfaces like gRPC / gNMI / OpenConfig. Another good example is to connect the dots between Application Experience problems and Network issues, requires a robust collection and analysis capability for flow data – a difficult to develop capability that requires a network focus. Even log analysis requires a network focus.
Network AIOps focuses on Network use cases. That focus drives development of the necessary interfaces, constructs, equipment types, suppliers, partnerships, algorithms, models, and types of anomalies.
Conclusion
Network Operations teams have multiple significant Network Tool Complaints. Addressing these complaints requires a clean page AI/ML, cloud-native approach that delivers significant improvements well beyond monitoring and observability. This is the vision of Augtera’s Network AI platform/service, to go beyond reactive Network Operations, and create a new era of preventative capabilities, so incidents can be remediated before they even occur, where Network Operations responses are automated, and where simplicity can be returned to Network Operations by integrating all Network Operations data into one anomaly detection and incident root identification platform. This is made possible, today, because of the general progress made by all of IT in AI/ML, and by purpose-built solutions for Network Operations.
