Can emerging technologies create self-healing networks?
When talking about Machine Learning (ML) and Artificial Intelligence (AI), it’s often forgotten to discuss the value of the data fed into it. How critical that data is, in shaping how it understands and reacts to information later on. How it ‘learns.’ There are many articles, books, and white papers worth investigating to understand this risk; from simply bad decisions, to creating a bias solutions.
During TechFieldDayX at Cisco Live EMEA this spring, JP Vasseur’s presentation on how they embeds ML/AI into networking, learning from the past, but being clear on the limitations. ML/AI isn’t the right use case for everything, for some situations it could even make things worse - such as environments that are not well understood or documented. The goal JP had with his team was to implement within the right use cases, where it can make a positive difference, and be scalable.
At the moment, we have a great opportunity to make use of existing technologies to see what’s worked and what’s not working. Using historic data fed into the tooling, we can try to estimate on the potential future of the application and network connectivity. Whilst we aren’t at a place where ML/AI is able to fully predict the best outcome, the focus is providing advisory on what actions might be worth taking.
JP highlighted we don’t have the confidence to let ML/AI completely lead, but by looking at the data, make predictions on what may happen.
“The goal is to say for some issues, can I predict the issues before they happen? If I cannot, fall back to reactive solutions” - JP Vasseur
The mindset to acknowledge the limitations, whilst still using what we have to improve our solutions. I wish more technologists looked at things in this way, as so many solutions are sold via marketing terms and stating they cover everything. But in reality, the gaps become glaring when deployed in the wrong scenario. When I was a consultant, I would work with companies to identify what they currently have, and at times, I found the solution in place didn’t fit their use case. That’s an expensive resolution.
The current state where JP demonstrated at Cisco Live EMEA, is to predict both physical failures, but also application failures, and switching paths before the issues cause an impact - to provide a smooth consumer experience, and providing as transparent as possible solution.
“When making predictions, from time to time, we are wrong. Therefore, if we make a recommendation, it must be right all the time - this means ultimately, you would be making less predictions. However, the ones that are predicted, are right.” - JP Vasseur
Use Cases where this is expected to be beneficial:
Predict network connections, to select the fastest path.
Identify application distributions before they cause delays.
Where it wants to go:
Use of predictive technologies to recommend changes, whilst still keeping the reactive in case things change.
To support high availability, reducing connectivity failures & enhancing application experience.
Enabling self healing networks with trusted automation.
Whilst it may not be there just yet, JP’s team are laying the ground work to create self-healing networks, that provide trusted automation. The mindset of “If we make a recommendation, it must be right or better,” sets the tone that when the solution is mature it will trusted to enhance our technology experience without impacting our availability.