putilov_denis - stock.adobe.com

Extreme Connect 2024: AI in networks to live or die by trust

Cloud networking provider unveils hub for research, development and innovation in networking previews, tapping AI to offer a new way to design, optimise and deploy networks

Over the past five years or so, artificial intelligence (AI) has proliferated throughout the world of networking driven by the complexity and scale of infrastructures, emerging as a transformative technology in optimising networks, and enhancing efficiency and reliability. However, according to Eduardo Kassner, chief data officer and general manager for data and AI at Microsoft, AI in networking will live or die on its basic usefulness in performing core and fundamental tasks, and will fail on everything else otherwise.

Introducing a keynote at Extreme Connect 2024, Extreme Networks chief product and technology officer Nabil Bukhari said that in networking, the possibilities with AI are limitless, yet that typically results in two emotions in equal parts: excitement and fear.

“When it is so exciting that you can do anything with it, it’s [also] scary that you can do anything with it,” he said.

“At Extreme, we believe in innovating in technology, and bringing these technologies in practical, thoughtful and secure ways so you can actually use them now for use cases, for the challenges that you face today, and be able to bring value to teams, companies and your customers downstream,” said Bukhari.

For Microsoft’s Kassner, the fear and excitement issue leads to trust. First, he noted that AI was not born yesterday, and that it has been around since the 1950s. However, he said many want to talk about it as if it’s new, and that AI deployment in networking is fundamentally expensive.

“You need a data scientist to train models,” he said. “You need a lot of compute. You need a lot of data. You need a lot of money. And then you pay the production and you need more money. [There are] many use cases that you will then spend [large] amounts of money on to use AI. And you need general AI and not just generative AI [GenAI].”

Addressing the question of trust, Kassner said everybody starts with the same five basic questions. “The first question is, is my data going to be private? The second is, am I going to be secure and compliant? The third is, if I’m successful [in AI deployment], and I do scale, how do I manage the scalability? The fourth, which could be interchangeable with the third, is how do I manage the cost, because this could get expensive. And then last, but not least, you need to be accurate.”

Adding value

Taking these into consideration, Kassner stressed the need for the AI setup to just work and add value to the job at hand no matter how many features it could support. Using an automotive analogy, he asked: “If you get into a truck, and you try to turn it on, and it doesn’t turn on, do you really care if it has a seatbelt? So, my point is simple. AI dies on it being useful and doing the purpose that you wanted it to do first, then [otherwise] it will fail on everything else. So, a lot of people are understanding that there’s a lot of scenarios where this is not going to be practical.

“But there are scenarios that are incredibly useful,” he said. “You’ve got to put the trust conversation into the piece that is useful to you, because you’re not going to build a computer language every time you build an application. You’re not going to create a database every time you need to store some data.

“And when you break it into scenarios, when you break it into a specific task, and you say, ‘This is what I’m going to use it for, and this is the benefit that I get out of it,’ then it’s a high-value scenario. And so that’s why we’re seeing [AI grow] faster than any other technology ... We can launch a system to production in four weeks, or five weeks. And it’d [offer] everything [such as] compliance [and be] secure, scalable, cost-effective and accurate. For us, every component of AI or GenAI falls within exactly the same rules that everything else does within Microsoft Azure. And that is the difference, so that we can build a system that complies across all of its components with privacy, sovereignty, compliance, etc.”

Stressing the vital concept of community, Eric McLaughlin, vice-president and general manager of the wireless solutions group at Intel, said it takes a village to build a successful AI application, and that a successful AI implementation for networking would require input from different partners.

“We come together with a myriad of components,” he said. “Most of us have some data right now, but I don’t have all the data; Microsoft doesn’t have all the data. You need data from different partners. You need processing at different places along the network, right from the cloud within the device. To make all of these things happen, and to actually deliver use cases that benefit users, we all have to work this together.

“So, what’s important? What value are you trying to drive? What’s your objective? What problem are you trying to solve? Start there, and then start building the value of that use case, the ROI of that use case, and then after that, the solution. Because if you don’t know what you’re trying to accomplish, and you don’t know the objective in doing it, you’re probably not going to get where you want to go anyway.”

Read more about AI in networking

As the keynote speakers were stressing the mission-critical need of trust in AI and the ecosystem, Extreme Networks announced the launch of Extreme Labs, an ecosystem with the stated aim of allowing creativity, collaboration and cutting-edge technology to converge to fuel innovation of early stage technologies, showcasing technology as it becomes closer to commercial availability.

And keeping with the theme of how the technology is driving networking innovation, the first technology preview will be Extreme AI Expert. Building on Extreme’s investment in AI beyond its ExtremeCloud IQ CoPilot AIOps offering, the technology comprises a GenAI service that is attributed with delivering substantial optimisations and cost savings in the design, deployment and management of enterprise networking and security.

Extreme added that unlike services that limit and silo data and knowledge, AI Expert wad designed to go beyond public product and network optimisation knowledge to deliver the best insights. It will combine public data with customers’ private data, securely, to create proactive recommendations that make it easier to accelerate the discovery of information and solve problems across a networking environment.

Benefits potentially include the ability to act as an expert on Extreme products and services to help work faster and smarter, accessing rich data from hundreds of thousands of source documents, translated into several languages. Data will be pulled from a combination of Extreme’s public repository, knowledge base and Global Technical Assistance Centre documentation.

AI Expert has also been built to pull data from the network and beyond to improve performance and operational efficiency, combining data from applications and devices across the network to establish intelligence on performance and experience. Extreme says the service will curate enterprise data to provide insights, automate operations and create alerts when it detects anomalies such as network overload, degradation or Wi-Fi dead spots, among others.

AI Expert is designed to turn insights into expertise and actions, recommending preventative actions and network optimisations based on business KPIs. Extreme creates suggestions and best practices to troubleshoot, resolve or proactively address issues.

AI Expert is currently at the tech preview stage, and Extreme expects to start integrating the technology into offerings later in 2024. “Collaborating with customers and partners throughout the innovation cycle gives them a voice and sets a bar of excellence, especially as we fully immerse ourselves in an era of AI, networking and security convergence,” said Bukhari.

Read more on Network security strategy

Data Center
Data Management