Creating a network capable of automatically responding to its own issues -- congestion, equipment failure, traffic spikes -- is one of the stated goals of today's virtualization push. Artificial intelligence and machine learning are key elements of this effort today, and they become even more critical going forward.
That's because the looming trends of 5G wireless and the Internet of Things will reshape networks and network traffic in even more unpredictable ways than are happening already, notes Mazin Gilbert, AVP of Intelligent Services at AT&T Labs. In this second of two stories on artificial intelligence (AI) and machine learning in telecom (you can read the first one here), we take a closer look at how AT&T Inc. (NYSE: T) and Level 3 Communications Inc. (NYSE: LVLT) are deploying technology today to be ready for future services.
First, there is a massive spike in data traffic expected with the advent of 5G and IoT, as much as a 10-fold increase in the next four years, Gilbert notes. In addition, 5G technology, with its millimeter wavelengths traveling shorter distances, will create many more network "edges" even as the IoT connects billions more devices from which data can be collected and analyzed.
"Things happen at the microsecond level and we have to react to that at a microsecond level and give the business or the consumer the right experience without any degradation to that traffic," Gilbert says. "That is really the challenge here and that is what we have been doing."
Already, telecom service providers are well aware that they are vulnerable to the next big "thing" that will hit the network on the timetable of its developer, not any service provider. "We don't have a spreadsheet for this, no one calls us to tell us they are going to be viewing videos heavily in the next few hours, or that business X is about to announce something that is going to flood traffic," he notes.
That's why tying AI and machine learning into the software-defined network control layer is critical to enable the network to respond faster than humans can to those unpredictable spikes. It starts with human expertise, Gilbert stresses, because the humans need to design the network capabilities to respond, but, in their responses to issues, they essentially "teach" the network how to respond to similar circumstances going forward.
"The resources are always going to be limited, the bandwidth is going to be limited," he says. "The ability to smartly and intelligently figure out how to move traffic from one place to the other is a big challenge. And there is tremendous room for AI and machine learning to do a better and a smarter job as we go to 5G, and anticipate traffic, not just react."
Level 3's SDN-based Adaptive Network Control goes one step further in trying to put automated tools into the hands of its enterprise customers, so they can support their hybrid cloud environments more dynamically. Today, most of what's possible falls into the first two categories of Level 3's internal branding for AI/machine learning of "See, Know, Act."
"We put into the hands of our customers the capability for them to help define what they want to know about and set it up as an alert," says Travis Ewert, VP of network software development. "The control piece -- that is the 'Act' portion -- that is where it gets really cool and borders on true intelligence. When you can connect that to your SDN control layer, now [network alerts] can prompt an automated network control or change event, then you've taken the human out."
He admits there is understandably a lot of caution around where that kind of automatic reaction is applied. Level 3 does also offer Enhanced Management analytics which, when paired with its Dynamic Capacity service, allows for programmable network triggers to scale bandwidth. That means its customers can custom-define network behavior based on analytics that they can prioritize. So, for example, the network can respond based on utilization, latency or some other factor, in a dynamic way.
"Today, for the greater part of the network stuff, we will use data to inform us as an issue is developing, and we still use our highly skilled technicians to then go manage that thereafter," Ewert says. "There is plenty we are working on that would even be more internally facing versus what we are already doing with customer external options."
And there are business rules and policies that act as safeguards for any automation, he says. But because Level 3 is already collecting massive volumes of data about its network for constant analysis, there is no need to do as much testing as part of the troubleshooting process -- anomalies are easier to detect and can identify the trouble spot more rapidly.
"The other thing that we are working on would be tying threat intelligence into the SDN control layer where it is not just auto-detect but auto-mitigate," Ewert says. "For that, we have a lot more defined use cases that we are working on and have a number in pilot right now."
In those scenarios, network functions virtualization plays a role, enabling a mitigation to be spun up on a virtual machine to provide IP-address blocking, traffic scrubbing for DDoS mitigation or a new firewall application, he adds. Those solutions are possible in hardware, but NFV enables greater speed and flexibility of deployment.
There is much more in the works for AI and machine learning at both companies, and much yet to be discovered as they heavily invest in these possibilities -- along with many of their telecom brethren.
"It is a major effort we are doing today -- I cannot tell you we have solved everything," Gilbert says. "We have put a lot of intelligence in our network today and as we get to 5G, there will be more and more technologies coming up in our network that help us alleviate and improve the experience."
— Carol Wilson, Editor-at-Large, Light Reading