Automation and Machine Learning /Artificial Intelligence in a Multi Vendor Reality

February 19, 2020 | Andre Pech, AVP Software Engineering, Arista

Learn More

Automation and machine learning / AI for networks are all the rage today; it’s a key aspiration for most large-scale organizations. But there’s a lot of hype out there.  The joke is “If it’s ML, it’s Python.  If it’s AI, Powerpoint”.  It’s difficult enough to apply these technologies across a single equipment/vendor domain but becomes daunting when trying to apply in a multi-vendor implementation.  So how do you build a foundation that supports these efforts?  What are some of the important architectural considerations?  In this session, you will learn how some of the biggest hyper-scale cloud companies have tackled this and learn about the lessons that Arista has gleaned in its journey to promote a true multi-vendor management plane.

We’ll explore the concept of Infrastructure as Code and how to leverage common industry tools effectively across vendor-specific domains.  We’ll discuss the importance of “state” when thinking about automation and machine learning, and review some examples of what’s possible if you do have the right state information collected in one place. We’ll also explore new standardization efforts around telemetry and automation through technologies like OpenConfig.  Finally, we’ll discuss what we believe to be a key requirement across all vendors in the infrastructure market: the creation of a federated model to share state and enable end-to-end correlation, baselining, anomaly detection and predictive analytics.