Enterprises see the potential for AI to benefit network management, but progress so far is limited by AI’s ability to work with company-specific network data and the range of devices that AI can see.
This is the time of year when enterprises look ahead to the new technology options they’ll consider for purchase in the coming year. They’ll usually focus on areas where they believe new technology will be most useful, where perhaps it’s the most needed. One such area is network management. Of 178 enterprise IT leaders who commented to me on the topic, 169 said they are looking for new network management options. Sounds great – but consider that only 37 said they have identified anything compelling so far. You have to wonder whether vendors have lost the map that leads to network management innovation and utility.
AI is the most obvious thing that could advance network management, but so far, it’s been less than overwhelming to enterprises. All 169 said they are interested in AI for network management, and 30 said that they have identified at least one option. Only 9 said that they have a budget for their identified option in 2024, and all of them reported a common problem with AI in network management—the problem of scope. Here, they mean two kinds of scope: the extent to which AI is specialized to their own network data, and the range of devices that AI can actually see.
The problem of data specialization is being handled by a combination of specialized AI/ML that operates explicitly on enterprises’ own network data, and by the extension of large-language-model tools to incorporate all manner of company data, including network telemetry. Users think more progress could be made here, but they also think AI is moving in the right direction. Most don’t think it will be there in 2024.
The scope of supported devices is another matter. The benefit of AI in network management is greater if AI can see all of the network; narrow-scope AI risks requiring operations personnel and AI elements to share management responsibility, and enterprise operations specialists say shared responsibility really means nobody is responsible.
But broad-scope tools are rare, and even more rarely integrated with management systems overall. Multi-vendor networks in particular are unlikely to be supported by integrated AI, and almost half of the enterprises with an interest in AI operations said they have at least two vendors in their networks. This is likely to take longer to fix than the data-scope problem, so enterprises don’t expect major gains next year.
AI takes on network management via application QoE
Well, it’s clear that enterprises think AI in network management is progressing, but it still needs some new kicker that’s a work in progress. Do enterprises see any areas of potential revolutionary progress in network management? Two main trends have their attention.
The first of these is a management take on something that’s already becoming visible in a broader way: absorbing the network into something else. Companies have said for years that the data center network, the LAN, is really driven by data center hardware/software planning and not by network planning. They’re now finding that a broader use of hybrid cloud, where the cloud becomes the front-end technology for application access, is pulling the WAN inside the cloud. The network, then, is becoming less visible, and thus explicit network management is becoming less important.
Ironically, this magnifies the potential role for AI. With everything getting subducted under the high-level view of sustaining application quality of experience (QoE), there are too many moving parts in an experience to allow for meaningful problem identification, isolation, and resolution. Enterprises would love to have a true “QoE console,” and they believe AI is the tool most likely to make that wish come true. Progress is visible here, enterprises say, but some vendor is going to have to take the lead to transform their options. They don’t believe a network vendor is likely to be the first mover.
Digital twins plus AI could broaden network management scope
The second development gaining attention is being proposed by a number of vendors, the largest being Nokia. It envisions using “digital twin” technology, something most commonly associated with IoT and industrial metaverse applications, to construct a software model of the network based on digital twins of network devices. With this approach, the network becomes in effect an industrial system, and potentially could then be monitored and controlled by tools designed for industrial IoT and industrial metaverse deployments.
One problem is a lack of support from vendors that enterprises know. Vendors that support digital twin modeling, like Forward Networks, have little name recognition among enterprise network planners. Only 29 of the 169 enterprises looking for new visions in network operations knew about Nokia’s ideas, partly because they had no engagement with Nokia to speak of, and partly because the new approach is part of Nokia’s Technology 2030 initiative, which was only announced at the end of October. If the digital-twin concept develops, it could reinforce AI too.
AI is typically viewed as a means of better exploiting information about network behavior, using it to discover issues, isolate problems, and recommend solutions. One challenge to that approach is the question of context or state. What is normal operation? How are the elements of a network supposed to relate to one another? What happens if we apply this suggested remedy? The answers to those questions might be discoverable in machine learning, but how long would it take for AI to learn enough about network operations to dig them out?
Digital twinning of a network could create a model that describes the relationship between elements and not just the state of the elements. Because that model represents an operating state, you could define multiple models to represent various acceptable states and expected error states. You could even run simulations through the model to see what outcomes would be likely before you took a step to change device or network behavior. Add AI to this, and you’d have an AI management tool that could not only see what’s going on, but also see what should be, and what might happen. It’s hard not to see that as a powerful path toward addressing both those network management scope issues.
Well, maybe not too hard to see, because of those 178 enterprise IT leaders, a grand total of four expressed an understanding of the potential relationship between AI and digital twinning. All of these were in the group who had heard of Nokia’s digital-twin approach, which suggests that enterprises think of network management advances in terms of what’s offered to them by vendors or written about in tech media. They don’t rush out to promote ideas; they exploit ideas.
That’s potentially bad, because generally vendors are looking to enterprises to offer perspectives on what new network management features might be valuable. Right now, the vendors are pushing AI because…well…competitors are pushing it and it’s being written about. Nokia has digital twin aspirations, but the company has not released enough of the details to know where it’s going and when it might get there. That might be never, if other vendors don’t take up the torch too.
To really change the game, network management can’t be just fielding alerts. That’s the lesson enterprises are confident in drawing from today’s offerings. As to what network management should be now, and what might get it to that point, enterprises are still on the fence. Maybe 2024 will bring clarity.
