As more and more devices come online, IT professionals working to maintain high-performing, secure networks will come to rely on the help of artificial intelligence technologies.
This article originally appeared on VMBlog.
During his keynote address at Cisco Live 2017 in June, Cisco CEO Chuck Robbins remarked, “We’re moving into a world of unbelievably massive expansiveness. Distributed connectivity across hundreds of billions of devices. And through artificial intelligence, through machine learning and scale, we have the ability to extract greater insights from all these connections than we ever have in the past.”
As Robbins’ highlighted, there are already 8.4 billion internet-connected devices in use worldwide, and experts predict that by 2020, as many as one million new network connections will be added every hour. Designing, deploying, managing, and optimizing networks capable of delivering reliable and secure internet access to that volume of devices is no small task.
But observers have high hopes for AI as a solution to exactly this problem, believing it to have the potential to help IT professionals optimize network performance regardless of the size of their user base.
“Machine learning” is a specific class of AI that focuses on giving computers the capacity to learn and produce insights in a manner that exceeds their explicit programming. By grouping, classifying, and analyzing vast amounts of data, a machine learning algorithm is able to pinpoint increasingly minute trends within a dataset — eventually picking out trends so nuanced or “hidden” that they are imperceptible to human observers.
Machine learning algorithms are already being used by a wide variety of businesses that consumers interact with on a daily basis. Users take suggestions from machine learning algorithms on Netflix and Amazon, which analyze millions of data points describing the shopping or viewing activity of millions of users to draw out larger behavioral patterns. Only an algorithm could develop these kinds of valuable insights, as it’s essentially impossible for a human data analyst to consistently notice correlations between near-infinite sets of data.
This kind of predictive, “intelligent” data processing is incredibly valuable in the networking space, especially since the data that IT professionals receive from their networks is often unlabeled, adding another layer of time-consuming work for analysts. A machine learning algorithm can use clustering to identify relationships between unlabeled data and build an analytical framework it can use to create new information and make critical predictions.
In practice, this means several things. For one, it makes wide area network (WAN) optimization far easier. Typically, dynamic routing through a WAN is achieved by establishing specific protocols that redirect traffic in the event of a link loss or similarly distinct occurrence. These path-selection protocols are helpful, to be sure, but they don’t offer network administrators much in terms of predicting bottlenecks or optimization their network.
What a machine learning algorithm can do is collect, organize, and analyze network data and, over time, use the resulting insights to create a baseline model of the network’s traffic. This model can then be used to determine which network path would be best for what traffic at any given time, when techniques like deduplication or compression should be deployed, and so on.
The bigger picture view is that machine learning can be used to gather, process, and act on information — including interface statistics, logged-in users, and longitudinal bandwidth consumption — about an organization’s entire network infrastructure. Once a machine learning algorithm is introduced to an organization’s entire network, it begins to assemble a comprehensive picture of it and pushes configurations to each component that, if applied, will guarantee optimal performance.
So what does the arrival of machine learning to the practice of networking mean for IT professionals? Fortunately, though AI technologies like machine learning do have the power to revolutionize network performance optimization, they can’t do so alone.
In the future — and to a large extent, right now — AI will reveal new ways of designing and managing faster, more cost-effective, and more secure networks. But IT professionals will still need to build the networks, troubleshoot critical issues with right-sized solutions, and manage the machine learning algorithms themselves.
Ultimately, as Robbins pointed out, the future of networking depends on our ability to combine the massive analytical power of AI with the creative and problem-solving capacities of human beings. If we are able to do this effectively, we will not only be well-prepared for the deluge of devices that will come online over the course of the next several years, but we will be able to deliver better, safer internet access than ever before.