Machine learning and artificial intelligence have gained notoriety among the general public through applications such as Siri, Alexa or Google Now. But, beyond consumer applications, these new hot areas of innovation are bringing unbelievable benefits to the different components of IT infrastructure that enable it, said David Meyer, Chairman of the Board at OpenDaylight, a Collaborative Project at The Linux Foundation, in his presentation at the DevOps Networking Forum last month.
This is most evident in compute technology, which grew more and more powerful to make machine learning and AI happen. It’s less obvious how networking contributes to and benefits from AI, he said. It’s still a new and growing area for networking — but it’s coming.
“I’ve never seen a technology this powerful that’s moving as fast or is as cool,” Meyer said. “Everybody I know who’s in networking or any other discipline when they understand what this is, they want to work on it.”
Meyer mentioned a talk on the future of telemetry at the NANOG 64 event that included machine learning as one example that the networking industry is starting to pay attention to it. The Internet Research Task Force (IRTF) is starting a working group on machine learning in networking, he said. And OPNFV’s failure prediction project is also a foray into machine learning.
It’s still unclear, however, how these and other networking projects will actually incorporate machine learning and AI. One among the many ideas that Meyer shared on stage was a huge sensor network that learns from previous experiences and predicts future outcomes.
”So what if you could do something like say, “Hey, this interface just flapped. We know from experience that this will cause, maybe, blackholing of some traffic or something,” Meyer said. “It depends on what’s in your data to know what you can learn and what you can predict, but this is the kind of capability we want to have and we’ve done some of this already.”
The potential applications for the networking industry are extremely powerful, he said.
“Remember that unlike statistical models that people are using, these things learn a computation, they learn a function,” he said. “We’re going to see more emphasis on real-time, control, and online learning, and reinforcement learning, and things like that.”
And it may be coming sooner than you think.
“All this stuff seems exotic maybe today,” Meyer said. “In a month or two, three months, four months, that kind of time frame, it’s going to start popping up in your space.”
Watch Meyer’s full presentation on the intersection of machine learning, networking, and DevOps, below.