Most network managers would agree that if a network were to deal with the more mundane matters of management itself, they would be overjoyed. What if this was more than just a fever dream of an overworked network admin’s brain? What if there was a way to teach networks how to do basic tasks and to inform the admins if more complex tasks requiring higher logic were required?
That would certainly be something to see. The reality of the situation is that we are already living parts of this dream. Machine learning (ML) has taken this dream out of the realm of impossibility and instead allowed us to see this implemented as a reality in our lifetime. The future has already started under our very noses.
Machine learning and artificial intelligence
One of the hottest topics of conversation in the world of technology these days is ML. SAS defines ML as a special branch of artificial intelligence (AI) that is built on the idea that systems can identify patterns and make decisions based on those patterns with very little human intervention. Why ML is important stems from how it sets the stage for everything we're expecting in our fantastic, fictional world where networks learn how to manage themselves. ML is the engine that drives the AI of the system. Thanks to the ability to learn from its decisions based on small amounts of human input, a computer can construct from each iteration become "smarter" with each success or failure and building up a view of how it should be operating. The evolution we're concerned with is going from a reactive system to a proactive system and then finally to an assurance system.
Understanding reactive systems
Currently there are already companies engaged in developing systems that operate on a rule-based system, allowing the network to do certain things if certain criteria are met. Oracle states that a reactive system is one that adapts to changes in the network condition, but which does so without needing manual intervention. Rules are important in setting flags for humans to investigate suspicious activity, but by themselves they don't afford a network the ability to actually be taught anything that requires higher thought processing. What we need in such a case is proactive network management.
Visit Innovation Enterprise's Machine Learning Innovation Summit, part DATAx New York on December 12–13, 2018
Evolving into proactive systems
With networks generating such a massive amount of data, storing them in a central repository is the best way to deal with this data. However, collecting the data is only the first step and all the other steps depend on how the data is processed and analyzed to offer insights. The same can be said of data generated by a network. By determining what the average condition of the network should be, we can then decide where problems may be occurring. As Hood et. al. note in their paper, if the system is given a normal operating state, then it can theoretically determine unknown or unseen faults based on data generated by the network itself. The idea of a network being able to report that it is malfunctioning is one thing, but being able to pinpoint what's causing the malfunction is a whole different order of significance. As Andrew Miller, former CEO of Polycom, has argued, this is the start of evolution from simply a connected series of computers into something that is seamlessly adaptive and can deal with most of its own problems.
Assurance - the end goal
Once we've set up a system for collection and analysis of network data, it's not a stretch to think that this data can be used for more than just determining faults on a network. With proper analysis, this data can create a lot more value by developing a means to deal with abnormal circumstances - sort of like a network manager does currently. Search Networking has noted that Cisco has already started development on an assurance engine that can be used to teach a network about itself assurance management will allow the AI to learn from its own data and adapt automatically to suit the network state. theoretically, it can even notice irregular network intrusions and move to combat it with network security directly.
A new network management paradigm
There is always going to be a need for network managers in any sort of network, but their job descriptions are likely to change as assurance networks start coming into their own. ML enables networks to spot problems as they occur and take measures in order to bring the network back into a stable, normal operating state. While this is currently still the realm of science fiction, the future looks very bright for the creation, evolution and implementation of assurance network management in the future.