Automation, Big Data and AI
The final topic David Gee and Christoph Jaggi mentioned in their interview was big data and AI (see also: automated workflows, hygiene of network automation and network automation security):
Two other concurrent buzzwords are big data and artificial intelligence. Can they be helpful for automation?
Big Data can provide a rich pool of event-sourcing information and, as infrastructures get more complex, it’s essential that automation triggers are as accurate as possible.
Big Data lends capabilities to use rich and high integrity data to source information for triggering and providing input to workflows.
Artificial Intelligence, Machine Learning and Deep Learning are all fantastic technologies when coupled with automation. Machine Learning is one of the simplest to consume, and telemetry type products make use of ML for the prediction of values. This is useful for preventing congestion in high traffic Internet events or predicting failures and minimizing business impact when they happen. Designers that couple machine-learning and event-sourcing gain powerful capabilities and uncover new areas where automation can make a real difference.
You know I’m not as optimistic as David is and tend to side more with James Mickens’ view. For a more down-to-earth approach check out what David had to say on event-driven automation in Spring 2018 Building Network Automation Solutions online course – you’ll get access to those materials as soon as you register for the course.
To avoid congestion we need to add extra BW at the "predicted" time (re-route traffic through other links?).
If we have "magically" gained knowledge/prediction (through any sort of AI) we need to act upon this knowledge i.e. add new BW to mitigate the congestion.
So we need to have this extra BW anyway.
It's still a mystery for me...
Obviously the challenge is how to figure out what the "usual" state is and that's where machine learning could be helpful.
Automated actions based on ML results is a totally different story. I wouldn't be as brave as that yet... but it all depends on how much you care about false positives and false negatives.
Of course we should know what anomaly is normal in the network to not be misled when real issue happens - to avoid investigation of anomalies which were "normal". I would like to avoid discussion why certain anomalies must retain in the network...