Machine learning in network management has promise, challenges

As part of the trend toward more automation and intelligence in enterprise networks, artificial intelligence and machine learning are increasingly in-demand because the ability to programmatically identify problems with the network and provide instant diagnosis of complex problems is a powerful one.

Applying AI and ML to network management can enable the consolidation of input from multiple management platforms for central analysis. Rather than IT staff manually combing through reports from diverse devices and applications, machine learning can make quick, automated diagnoses of problems.

Gartner senior director and analyst Josh Chessman laid out the problem for the IT worker that machine learning is designed to solve: “I’ve got all these monitoring tools, and they’re all telling me something’s wrong, but they’re not telling me where it is. The biggest strength with this stuff today is that it can identify ‘you’ve got 26 events from seven different tools, and they’re all about a network problem.’”

It’s difficult to say how rapidly enterprises are buying AI and ML systems, but analysts say adoption is in the early stages.

One sticking point is confusion about what, exactly, AI and ML mean. Those imagining AI as being able to effortlessly identify attempted intruders, and to analyze and optimize traffic flows will be disappointed. The use of the term AI to describe what’s really happening with new network management tools is something of an overstatement, according to Mark Leary, research director at IDC.

“Vendors, when they talk about their AI/ML capabilities, if you get an honest read from them, they’re talking about machine learning, not AI,” he said.

There isn’t a hard-and-fast definitional split between the two terms. Broadly, they both describe the same concept—algorithms that can read data from multiple sources and adjust their outputs accordingly. AI is most accurately applied to more robust expressions of that idea than to a system that can identify the source of a specific problem in an enterprise computing network, according to experts.

“We’re probably overusing the term AI, because some of these things, like predictive maintenance, have been in the field for a while now,” said Jagjeet Gill, a principal in Deloitte’s strategy practice.

Another sticking point for a lot of ML systems is cross-compatibility. Much of what’s on the market currently takes the form of a vendor adding a new feature to one of its existing products. That’s handy for all-Cisco shops, for example, but can be a problem in a multi-vendor environment. “A lot of vendors are adding AIops because it’s kind of a buzzword,” said Chessman. “It doesn’t give you a lot of visibility into products from other vendors.”

There are vendor-agnostic ML systems for network management out there—Moogsoft and BigPanda are two of the bigger names in the field—but it’s more common to find ML features bundled with specific vendors’ products. “So take Netscout. They’ve got some ML, and it does a good job, but it’s focused on Netscout [products],” Chessman said.

Regardless of the hurdles the technology has to overcome, ML is likely to make many IT professionals’ jobs a lot easier, according to Peter Suh, the head of Accenture’s North American network practice. “Having those types of tools and solutions is going to be good,” he said. “It’ll help you walk through what’s going on on the network at any given time.”

While it’s also a potential step in the direction of full network automation, it might also result in the loss of jobs for IT staff, that’s not likely to happen in the immediate future, according to Gartner’s Chessman. What’s more probable is that ML will help free up IT staff to work on more revenue-generating activities, rather than putting out fires, he said. “Full automation is still years and years away.”