AT&T's Gilbert discusses how to engineer an autonomous network

Closed-loop automation will lead to a deeper application of artificial intelligence, according to AT&T's Mazin Gilbert. (AT&T)

AT&T's Mazin Gilbert has a lot of balls in the air when it comes to juggling the telco's virtualization and open source initiatives.

Gilbert, vice president of advanced technology and systems at AT&T Labs, is the technical steering committee (TSC) chair of the Open Network Automation Platform (ONAP). He's also involved with Acumos, which is hosted by the Linux Foundation's LF Deep Learning Foundation. Acumos was created to simplify the development of artificial intelligence while also providing a marketplace for AI apps.

Gilbert's work is on point for AT&T reaching its previously stated goal of having 75% of its network virtualized by 2020.

"More than 55% of our network functions were virtualized at the end of 2017, and we're well on our way to meet or exceed our goal of 75% virtualized by 2020," said AT&T Communications CEO John Donovan, according to a Seeking Alpha transcript of last week's second-quarter earnings call.

In this Q&A, which was lightly edited for clarity and length, Gilbert talks about how ONAP, machine learning, artificial intelligence and closed-loop automation are building toward a fully autonomous network.

FierceTelecom: The second software release for ONAP, which was called Beijing, was released recently. Can you talk about the most important elements of that release?

Mazin Gilbert: Number one is we have been investing significantly and heavily into the ONAP open source on every release. If you go and look at the statistics, we are pretty much the number one contributor for all the releases. We're very dedicated and committed to ONAP.

Number two is that we use ONAP entirely. As you know, we took components of our ECOMP and combined them with China Mobile's OPEN-O to create ONAP in open source. We finished last year with 55% of our virtualization target and we are expecting ourselves to reach an aggressive goal of 65% by the end of this year.

As we're doing that, we're moving towards more automation, more containerization, more orchestration and more closed-loop automation. It's not just meeting our virtualization target. Every release we've made, we are investing in the projects and we're deploying those projects in AT&T. We're building a community that we work with that really never existed as much before. Every release the success in the community is larger. The contributions aren't from AT&T only, they're coming from other operators and our vendors.

I think we have nearly 400 different committers in the Beijing release. These are not the architects; these are the software committers. That's pretty extensive when you look at what it was the release before that, which was half. It's increasing fairly rapidly release after release in that diversity.

I'm the chair of the ONAP TSC and my colleague, Chris Rice, is chair of the ONAP board. We have an obligation as a company to grow the community, to bring the operators all with us, to all move in the same direction, to help the operators to build up their skill sets, to build up and educate them to move in this direction faster. That is an obligation that we take very seriously as a company to bring the global partners and the companies all together to push in the same direction.

RELATED: MEF, TM Forum pitch in on ONAP's Beijing software release

FierceTelecom: Are there specific elements of the Beijing release that AT&T is interested in?

Gilbert: We are focusing on several fronts. One is that we're obviously focusing on driving our microservices architecture and using Kubernetes and other technologies to allow us to do the management and the orchestration. That's number one.

Number two, we believe that when the dust is settled, one of the key differentiators and value adds of ONAP is the data. It's the data that can be used for machine learning and closed-loop automation. Now you can drive value to add on top of the software. We continue to invest in all the relevant components to drive complete closed-loop automation. That includes the data collection itself, the data analytics, the machine learning and the policy. Defining the policy and defining the control both on the network layer control and the application layer control. Of course, it also includes monitoring these closed loops on an ongoing basis. We exposed some of that capability in Beijing with something called "CLAMP," which stands for Control Loop Automation Management Platform.

CLAMP is a project in Beijing that helps to design, instantiate and deploy closed loops. You no longer need to build one from scratch by integrating components. Now you have a design studio. You can start plugging things together and bang! Magic happens. That's CLAMP.

The last one is, but not the least, we obviously continue to invest heavily in the design of services and virtual network functions (VNFs). The third one is the entire VNF and NFVi (network functions virtualization infrastructure) ecosystem. That includes simplifying the onboarding of those VNFs, designing services by chaining different VNFs together, certifying VNFs and deploying those VNFs.

FierceTelecom: Once you've completed closed-loop automation, that's where you get more meaningful artificial intelligence?

Gilbert: Absolutely. You've got that exactly right. First, in ONAP, you have to build the foundation of a platform. We're doing that and we have two releases behind us. The second thing is that you need to enable this explosion of data to happen. Why? Because you need that for monitoring closed loop. That is where machine learning and AI play a big role. A closed loop is a perfect example of an autonomous artificial intelligence system. Complete. A lot of what we do in CLAMP and ONAP, in terms of data analytics and prediction, is the perfect example of machine learning. That's what machine learning is all about. We are now getting to this next level where we have some infrastructure in place. We have a community in place. We're now moving towards machine learning and AI.

This is why we formed Acumos under the Deep Learning Foundation umbrella a couple months ago. Now we're starting to get to that next generation of the network. I call it the autonomous network where now we bring you software-defined networking with machine learning and AI to truly create an autonomous network.