Previously published on SDxCentral
December 23rd 2019
While most of these technologies are still years from reaching maturity, that’s not stopping companies in the performance analytics space like EXFO and Vitria from investing big in machine learning (ML) and AI.
According to Ken Gold, director of test, monitoring, and analytics solutions at EXFO, the implicit complexity associated with massive 5G IoT deployments is only going to make identifying and resolving network anomalies all the more challenging.
Today, 38% of trouble tickets associated with network issues are raised by just 1% of customers, he explained. “If there are 100 [support] tickets you know about there are probably 4,000 that you don’t.”
For most customers, Gold says the occasional dropped call or service outage isn’t that big of a deal, but IoT deployments are less forgiving. As the technology marches toward large-scale adoption, the need to be able to accurately predict traffic and bandwidth demands is going to accelerate rapidly.
In what might sound like the plot of a “Terminator” movie, Gold says ML and later AI will play a big part in making this complexity manageable. He said EXFO sees ML being used to learn and identify what normal network traffic looks like to better identify the root cause of an issue when it crops up. Needless to say, our IoT devices probably won’t be turning against us anytime soon.
But while EXFO is preparing to launch an AI and ML-based AIOPs platform in the near future, other vendors like Masergy, Nyansa, and Vitria have already opened their arms to our AI overlords.
In fact, the degree of visibility made possible by AI has implications for more near-term applications than 5G and large scale industrial IoT including gains in operation efficiency, SD-WAN, broadband performance, compliance testing, and improved customer experiences.
According to Partho Mistra, VP of engineering at Cumulous, the application of AI to SDN is potentially huge.
He explained that every modern switch, router, or IoT device is capable of streaming telemetry data that can be viewed in real-time and analyzed. “The first application of that is to find things that aren’t working properly,” he said.
Mistra said that by using ML and AI it’s possible to catch incredibly subtle things that a human would never catch even if they stared at the data for days. And this is exactly what Masergy and Nyansa are trying to do with their AIOPs platforms.
Masergy earlier this year debuted its first AIOps platform with the goal of identifying opportunities for networking, security, and application optimization.
The managed SD-WAN service provider billed its AIOPs as a virtual engineer living inside its intelligent service control dashboard. In its initial release, the service used data that was mined using a combination of anomaly detection and predictive analytics to surface relevant data and recommendations for how customers can improve application performance, predict bandwidth needs, and optimize network throughput. The company says its AIOPs platform has the potential to significantly reduce downtime, enable faster fault recovery, and reduce the time to resolution for troubleshooting.
Similarly, Nyansa is attempting to use ML to consolidate networking telemetry from enterprise hardware that wouldn’t normally talk to each other. The company aims to make this telemetry more useful by making it easier for customers to connect information from one appliance to an event on another.
Nyansa’s Voyance AIOPs platform works by pulling in real-time and historical data from devices on the network. This information is then processed using a series of ML and AI algorithms designed specifically to solve network challenges. The company’s software automatically discovers and inventories switches and connected devices while monitoring their current state.
According to Mistra, the next intuitive step for AI in SDN is to close the loop and allow the AI to start acting on these anomalies.
“Can you go close the loop and go and self-correct the network?” he asked. “In a lot of cases, technically the answer is yes. … In practice, I think that’ll take at least 10 years for that to happen.”