SAN FRANCISCO—AT&T's Mazin Gilbert has a deep knowledge of the various open source communities that are in play across the telecom space, but he's also able to paint a big picture of how they will converge.
Gilbert, vice president of advanced technology and systems at AT&T Labs, outlined the elements that will lead to autonomous networks for service providers and the building blocks that are needed for next generation networks in previous Q&As with FierceTelecom.
During Monday's AT&T Spark conference in San Francisco, Gilbert spoke about how open source communities—both in and outside of the telco space—are coalescing.
On Monday, AT&T and Tech Mahindra announced the winner of their Acumos AI Challenge, an open source developer contest designed to spur the growth of artificial intelligence solutions. While AT&T and Gilbert have a vested interest in technologies such as 5G and the use of artificial intelligence, the contest was designed to go beyond the telecom industry, Gilbert said. Tech Mahindra and AT&T developed Acumos last year to drive the use and development of artificial intelligence.
Acumos is now part of the LF Deep Learning Foundation, which is an umbrella organization under the auspices of The Linux Foundation, that supports and sustains open source innovation in artificial intelligence, machine learning and deep learning
"We didn't focus on one industry," Gilbert said ahead of the contest winner being announced Monday afternoon. "We didn't say it was telecom. So you'll see that some of the submissions we had—healthcare, security, real estate, enterprise—were all over the place. It was purposely done that way to restart building up that foundation of the marketplace, because we need to drive the volume up.
"I think what you're going to see over time is many companies coming in and having new Acumos-like challenges in different industries and in this different vertical. We could have easily pushed this to be in the area of 5G, in the area of software-defined network or in the area of IoT, but we decided to really keep it open because it was really to try and drive an open source community around it."
In this Q&A, which was edited for clarity and length, Gilbert discusses the state of play for the various open source groups, and how projects such as Akraino will be more closely aligned with the Open Networking Foundation (ONAP.)
FierceTelecom: At the Spark panel, "What's next for open source?" there was some discussion about how open source groups both in and out of the Linux Foundation need to cooperate with each other. What are your thoughts?
Mazin Gilbert: Under the Linux Foundation, we've created what they call umbrella projects [which] have a lot of autonomy. There are multiple projects all connected over and over in the Linux Foundation Networking Umbrella. There are several projects which are very much connected with each other, like OPNFV and FD.io, ONAP and Tungsten Fabric. All these have used each other and depend on each other.
We've also created one called the LF Deep Learning Foundation umbrella. A deep learning umbrella is not much to do with the Linux Foundation Networking Umbrella. It's a separate umbrella focusing on deep learning, and machine learning and data science. And Acumos was the first project. We have two new projects that came in from Baidu and from Tencent. One is called Angel, and the other one is called EDL.
These projects are very much machine learning and AI focused. We're creating all these umbrellas that have tremendous dependencies within themselves. We think that Akraino will ultimately grow into its own umbrella for edge. At the moment it's a project, and you're going to see that growing over time.
What we're starting to do now is exploring the complimentary nature of these umbrellas. In ONAP, when you do closed-loop automation you need to do machine learning to predict failures in the network. Well, ONAP is not going to start the machine learning foundation. We're working now with the Deep Learning Foundation to create deeper integration. When I build a container that has some machine learning capability that predicts a fault of, let's say, a VM (virtual machine), then I submit that at the Acumos marketplace and it goes into the design studio for ONAP. When you design a service, that container is completely compatible with ONAP and you can build closed-loop automation with it. We're creating that deep sort of integration now between these umbrellas.
With Acumos, there's some machine learning that you'll see in there for AR/VR. So, Acumos and Deep Learning are not about telecom. It's broad. You're going to see some announcements made soon by some of our other companies in areas that have nothing to do with telecommunications. It was purposely done that way. From an AT&T perspective, we're going to create that link, between AI and 5G, 5G and Linux Foundation Networking ONAP. Between LFN and ONAP and Akraino. I call it the diamond.
FierceTelecom: What about other organizations such as ETSI? Where does it fit into that diamond?
Gilbert: I'm the chair of the ONAP TSC (technical steering committee) and we have a committee that's focused on SDOs (standards development organizations) and focused on external APIs. We have a project focused on that and a committee focused on that. I have a team focused on working with ETSI and at the moment what we onboard in the service designing creation in ONAP will support the ETSI APIs. And that's a key part because a lot of our members want to have full compliance with ETSI, ETSI APIs and ETSI standards. We're absolutely going to do that. Within ONAP we'll still have open APIs and some of those will be ETSI compliant and some of those will be broader than that because it depends on how the projects connect with each other.
FierceTelecom: Would an example of collaboration be how ONAP, MEF and the TM Forum worked together on MEF's northbound API in the ONAP Beijing software release?
Gilbert: Exactly. We work very closely with those folks. We work with the MEF. We work with ETSI. We are trying to establish other relationships with some of the other SDOs right now, and at some point I want to establish a relationship with oRAN on 5G, because our (ONAP software release) Casablanca is going to have some 5G use cases. We need to work with oRAN because they're defining the virtualization in the 5G architecture. I think we want to connect more with these communities and the power here is that network connections between those communities.
FierceTelecom: What about the Open Networking Foundation (ONF) and its work around CORD and edge compute?
Gilbert: We are also working with ONF, and I think you're going to see more with ONF, especially on the Akraino side. Akraino is all about creating these blueprints for edge compute and edge cloud, and you're going see a relationship now between ONF and Akraino very soon.
Remember, from an Akraino standpoint, there's no one-size-fits all for the edge. Even from an AT&T standpoint. Even though we will have hundreds and thousands of these edges, it's not a one-size-fits all. We may end up putting a few different technologies in some edges, versus another. Some edges may have to have high performance compute and GPUs. Some edges don't need to do that. So we're creating these blueprints, integration and testing in the Akraino community, and that's how ONF is going to be part of that. And so we will deploy some blueprints in some sites, and some other blueprints in very different sites.
FierceTelecom: Is there a timeline on these blueprints, or on the work that the open source groups are doing together?
Gilbert: I don't have a timeline to give you, but I'll give you our strategy of how we are rolling this out. The first thing you need to do is to get the foundation, and the foundation is the network workload. You have heard us talking about network clouds. You have heard us talking about Akraino and ONAP. We're going to use that to really push a lot of our blueprints to the edge. You have heard us talking about Airship as well at this conference.
The second part, you're going see us talking about 5G. We've made a number of 5G announcements today. You need the access. The network is one piece of the puzzle. You need the access on top of that to really provide that latency and that speed and make it ubiquitous whatever it is and drive a set of applications that you just can't do today. That is what we're starting to do that now. Our investment in oRAN, our investment in xRAN; you're going to see some announcement in that space.
Now as we are doing that, the third part of this movement is the applications. The non-network applications, but you need the foundation. You need the access. You need the network itself. You need it to be programmable and policy driven, so when you're driving these applications I can change bandwidth, I can change traffic whenever I want. We can't do that today. You have to have a programmable network, a policy driven network, a model-based network. You have to have access to the point where you can be interoperable. You can start virtualizing part of this access to make it programmable. This is the strategy we are pushing very hard on. As we are doing that, we are already testing out applications to understand "What are the requirements and the assumptions, for these applications requiring edge and 5G?"
What's the value of speed? Where do you need that and what kind of applications can we run at the edge that we can't do otherwise? Are these at 10 milliseconds, or 20 milliseconds, or 50 milliseconds? Makes all the difference. We've starting building these applications now, and testing them as we speak, and all of that is giving us the requirements of how we scale, really the edge moving forward. And that's exactly the sort of three-tiered strategy that I'm talking about.
FierceTelecom: All of which eventually leads to convergence of networks, services and applications?
Gilbert: Honestly, for me at least, that's the most exciting part. I know that we talk about, and people talk, about the edge as the hammer to solve all people's problems. They talk about AI. They talk about 5G. To me, the most exciting part is that convergence. It really is that convergence because ultimately every one of them brings a different value to the game.
I'll give you a simple example, even with the 5G and edge that we're doing, I can't scale 5G and edge. If I'm going to put thousands, tens of thousands of these locations out there, where do you put them? These are all small cells. Remember, microwave small cells. Where do you put them? To make sure that you get coverage, constantly, seamlessly, wherever you go. You're on the go, right? You're mobile. How are you going to get that?
Well that turns out to be an AI problem that we've solved for ourselves. Given everything we know—all the locations we have, and all the assets we have, all the township guidelines and requirements—where do you think our customers are going to be three, five years from now? Some customers will want more bandwidth than others. Not all of us are the same, right? Putting all that together to try and say "Okay, we're announcing in this corner of Dallas." And then when we announce it; this is not magic. This is all technology, and AI is driving it. There are reasons for that. It's because of how we see all these drivers coming in together to tell us this is where you need to build up first. Ultimately, we will all have 5G many years from now ubiquitously across the United States. How you get there, how you build it and how you bring in these other complimentary capabilities we talked about, ONAP, Akraino, Acumos, edge and AI, is what is exciting to me.