HPE Aruba Networking Blogs

The secrets of AI networking: Part 1

By Jose Tellado, VP of AIOps, HPE Aruba Networking

Was it the “60 Minutes” episode that aired last year, or perhaps the salvo of coverage unleashed by the Wall Street Journal, CNBC, and other business media starting last spring? The interest in artificial intelligence spurred by ChatGPT and generative AI is beyond palpable for me not just at work, but at the kitchen table.

As the lead data scientist at HPE Aruba Networking, I’ve been inundated in the past 12 months by customers, colleagues, and friends with questions about the impact of AI on networking: Why is it necessary? Does it work? Does it work across all types of environments? What are the essential ingredients of successful AI networking?

As enterprise networks grow more diverse and distributed, AI technologies are becoming critical to the network management process—since it is increasingly impractical to manage and protect these environments with traditional manual techniques. Planning, installing, securing, and operating myriad access points, switches, gateways, and WAN connectivity require the insight and efficiency that AI can provide. CLI, while still alive, is on life support.

But you have to look beyond the headlines to understand if a particular solution will actually solve problems, such as too many trouble tickets, elongated investigations, and security blind spots. In this blog series, I’m going to “double-click” on AI networking to help you evaluate AI claims and more importantly, AI effectiveness.

I’ll start with examining the technology trends that had to come together to make the application of AI techniques to network management practical:

  • Data science combined with domain expertise. The first breakthrough was the growth of data scientists who studied how networks operate in order to train, validate, and deploy AI models and techniques to solve real-world problems. I’ve had the privilege to lead such a team in recent years. What our scientists have learned is producing customer-facing, practical results with AI involves a lot more than standing up a few models with a small amount of data. AI effectiveness comes from continuous iteration on large, rich, and diverse data sets—implementing a model and monitoring and tuning it to optimize results in all customer environments. This AI effectiveness, in conjunction with programmable, flexible, and proven network hardware and software designs (whether it is advanced RF designs for APs or microservices operating system architectures), ultimately improves network performance and reliability—for both the network admin and user experience.
  • The right data. This is where our data scientists, device software engineers, and network and security engineers work together to extract the telemetry needed to train, validate, monitor, and continually tune and train the AI models. Anyone can say “we collect huge volumes of data,” or if they don’t have enough data, “we collect the right data,” but the proof is in the usefulness of the results. As generative AI has proven, you need the right data at very high volumes, with that data generated from a broad and diverse set of environments.
  • Cloud control planes that scale. Finally, it is the move of the network control plane to the cloud that makes this all possible. This provides a central location to collect the necessary AI training data while providing the single point of visibility and control to see and implement the recommended AI actions, which could be applied in the cloud or pushed to the devices. But, not all clouds are equal. Most are built for smaller environments and lack support for large numbers of connections, seamless roaming, security orchestration, etc. So, while a cloud-based platform enables better AI, without a scalable and robust control plane, it won’t meet enterprise needs.

Like most technologies, cloud, data science and scale can be combined in a number of different ways depending on the target use cases, and it is exciting that network vendors are using AI to attack a wide range of challenges across all sizes of customer environments.  Over the next several blogs, I’ll explore what it means for AI networking to scale in the cloud, attack and solve the right network problems and reveal how a large and diverse data lake makes all the difference when it comes to a better running network.

We’re HPE and we have incredible AI resources to help customers both build their own AI solutions as well as incorporate AI in the products we sell—like our AI networking.

As part of this AI powerhouse, the secrets of AI networking aren’t really secrets if you know where to look. Learn more about our security-first, AI-powered networking solutions.