The topic of AI in the networking space has quickly moved from “what if” to “how soon”. Every major networking vendor is leveraging some aspect of AI in their products, and roadmaps list AI as a pivotal element for modernizing networks.
Revolutionary improvements driven by AIOps will significantly impact the way we operate modern networks. This is made possible by the contextually rich data available to feed AI/ML models in network observability solutions, and the intelligence of those systems to produce outcomes.
Networking and AIOps
There are growing use cases for AI in the network, including, amongst others:
- Anomaly Detection
- Proactive Monitoring
- Capacity Planning
- Forecasting / Prediction
- Root Cause Identification
- Threat Detection
Further, I see AI becoming much more integrated into the network team itself – detecting and resolving issues, building and deploying configurations using intuitive chat prompts, and writing documentation. AI will effectively have a seat at the table for architecture and engineering decisions.
This is early days but we’re already starting to see the real-world possibilities with solution providers like Selector.AI, who has developed an AIOps platform to tackle network challenges in the modern age.
What is Selector doing?
The use of AI/ML algorithms is now an important consideration when choosing network observability solutions. This is because the existing approach of operating a networks has a few major challenges:
- Heterogeneous data is spread across many domains and dashboards
- Manual analysis is needed to correlate events between systems, dashboards, teams
- Siloed Teams do not collaborate effectively, nor do they understand each other’s metrics and operational indicators
Selector overcomes these challenges with what the call the three “C’s” – Collect, Correlate, Collaborate.
Selector‘s main product, Selector Analytics, uses neural networks with mathematical models to analyze data, learn patterns, and intelligently report on information to significantly improve operations by pointing specifically to network issues and their likely root cause.
To achieve this, Selector Analytics provides a high volume data ingest architecture for scale out via k8s, data distribution event streaming via Kafka, and disaggregated collectors for streaming data. Some supported connectors include:
- Protocols such as SNMP and streaming telemetry (e.g., gNMI)
- Event technologies such as Kafka and RabbitMQ
- Data stores such as Prometheus, influxdb, and Splunk
- Cloud providers such as GCP, AWS, Azure
Many of their customers are already ingesting into a Kafka bus, simplifying Selector’s ingestion method. They advertise minimal time to integrate new connectors using their automated compiler.
ML converts messages received via the connectors into a visual “red/green” logic based on the trained models, with successive correlation performed between the reds, effectively optimizing the volume of data the software actually has to look at. Event correlation is enhanced using contextual tags attached to events and then comparing events with each other.
Using Named Entity Recognition as an ML technique, Selector is able to understand what is an IP address, what is an ASN, etc., from log messages that are evaluated, even if the structure of the log message changes. This is profound – no more parsing varying message structures with our caveman eyeballs or trying to write the perfect regular expression to find some unique value in a sea of symbols.
Uniquely, SelectorAI’s interactive interface is your enterprise chat app, Slack or Teams. Yes, there’s a UI for administration and deep technical analysis, but how you primarily interact with the AI is via Slack/Teams. It’s almost as though the AI is an engineer on your team. 👀
Selector uses Natural Language Generation derived from JSON objects of the correlated events, allowing for friendly communication with the AI.
Selector’s vision is that Enterprise communication tools are the way engineers will interact with the network data going forward. Moreover, since Selector use enterprise communication tools, this means they naturally have a “mobile app”.
Addressing those AIOps use cases
A demonstration was provided during a recent live streamed demo of the product, where CTO and co-founder Nitin Kumar used his phone to interact with the AI via Slack after receiving a message that a network issue had occurred. Based on correlated data, Nitin was informed that the cause of the issue was likely junior network engineer, Joe.
This is the bread and butter of Selector Analytics – providing rich and accurate root cause identification. But that’s not all for Selector – forecasting and predictive analytics is in the roadmap as one of their top priorities.
Remember, Selector functions by analyzing massive amounts of network data to identify patterns and trends. As the platform gets smarter, predictive models will be used not just for identifying issues, but for forecasting future network behavior, growth demands, and potential network optimizations based on historical insights into correlated network patterns.
How can it be deployed – how’s it licensed?
Many customers do not want to send their logs to the cloud – so they offer multiple deployment options, including on-prem, cloud, SaaS.
Predictable pricing is based on the number of monitored devices rather than the volume of data. This is the opposite licensing model for most other solutions, such as Splunk, which price based on data volume.
There’s much to unfold with Selector – and AIOps in general. With anything new, trust must be earned, especially so when it comes to AI. Will we get there? I think so. Purpose-built vendor-neutral solutions like Selector are setting the stage for what’s to come with network observability.
I don’t expect us to all become data scientists with the advent of AI in our networks, much as I don’t expect us to become developers with the adoption of automation. However, some knowledge about the capability with fundamental hands-on will keep you updated and prepared for the inevitable conversations ahead. It was clear to me after watching Selector’s presentation at Networking Field Day 30 that I have some self-education to do – so I’ll be experimenting further with AI tools and further educating myself on the ML concepts.
Stay tuned for more on Selector… and check out their presentation below.