Anschutz on AT&T's Use of Web-Scale IT

The principles of web-scale Internet companies -- including Google, Facebook, Amazon and Microsoft -- have filtered into the telecom sector with service providers such as AT&T.

While those web-scale pioneers made their mark largely in data centers, telcos have adopted web-scale operations and service for their own use in central offices and other areas such as virtualization.

According to Virtupedia, web-scale is defined as a global-class of computing, or architectural approach, used to deliver the capabilities of those large cloud service providers within an enterprise IT setting. The approach is to design, build and manage data center infrastructure where capabilities go beyond scale in terms of size to encompass scale as it pertains to speed and agility.

In Part I of this Q&A, AT&T's Tom Anschutz, distinguished member of the company's technical staff, talks about how it has modified some of the web-scale principles for its own use in areas such as virtualization. In Part II, Anschutz delves into how web-scale works across ONAP, CORD and AT&T's Indigo platform. (See Pacewicz: Indigo the Next Innovation Wave for AT&T.)

Telco Transformation: How do you define web-scale operations and services internally at AT&T Inc. (NYSE: T) , and why are they important?

Tom Anschutz: Wow, so that's a great question. Normally, web-scale services and operations are associated with these large big block data centers. And in some ways, the problem that AT&T faces is not that everything should be done homogeneously in one humongous place. But rather it's getting the same benefits, the same type of common methods and operations and common scale mechanisms across a large and distributed set of much more modest-sized data centers, or central offices, whatever you want to call it. The infrastructure kind of locations.

TT: Web-scale started out with Facebook, Google (Nasdaq: GOOG) and Amazon as a data center play at the beginning. They're viewed as pioneers in this area, but are these principles that AT&T was already working toward? If not, when did you start adopting them?

TA: So a lot of the principles were already in common. But, like most things, if you look at an adjacent industry, you learn from that. AT&T has adopted some of the new principles, especially those around what I would call "zero touch" or "lights out" type of on-boarding, and management of infrastructure.

And indeed, one of the things that we're working on this year that's really exciting is how we can make use of what I call "access to the source," which means both open source hardware and open source software, and make adjustments to facilitate what I call "operations simplicity," or make life simpler for the folks that manage and operate and fix problems in our network, and in our infrastructure. So it's pretty fun.

TA: Yeah, I can give you a specific example. So, in the past, we would have network equipment that was supplied by a broad variety of OEMs. And the typical model for that equipment would be that they would design and build something that they believed would address a very large market across multiple carriers. They would have their own sort of views around how it ought to be installed and managed and operated. They'd provide an element management system with that, which we would then spend a lot of time and money with our IT organization to integrate into AT&T's specific operations systems and tooling, and so forth. That's how it used to be.

The fun kind of work we have this year is that as we start developing more Open Compute Project [OCP] spec hardware, what we do is we learn, or we copy, or whatever you want call it. We give that engineering nod to what got developed at places like Amazon.com Inc. (Nasdaq: AMZN) and Facebook , and start looking at automating the on-boarding of equipment, and making it much more homogeneous across multiple different equipment types.

So, for example, as we work toward an AT&T-designed and specified XGS PON, or an optical access device, we're including things like ONIE, ONIE boot loaders and an ability to load an OS from zero. Also the ability to automatically inventory and attach these kinds of devices, using the kinds of tools that you might have seen formerly in large cloud data centers.

So in this new model, we no longer have what I would call this integration between an EMS in our operations systems. We have an ability to specify and then ingest equipment that has very common patterns for its management, on-boarding, loading images, taking inventory, managing faults and so forth. In fact, that can add up to real time and money savings for a company like AT&T in the long run.

TT: What about the use of web-scale operation and services for virtualization, such as VNFs, as well as prepping for 5G?

TA: So when you talk about web-scale, you said operations and services. In many cases, I would separate those two things. Web-scale operations are really an outgrowth or sort of a culmination of what we just finished discussing. That the way that the system is designed and put together has a greater amount of automation Things like self-learning, self-assembly and self what I would call, scale-in and scale-out characteristics.

Then there's a second part and that's sort of the service itself, or the workload, the composition on top of the infrastructure. In that case, you're right. We're very much looking toward new solutions that are less like what they used to be. They were vertically integrated or monolithic blocks of, bundles of, functions. We're looking to sort of tease apart more atomic, more reusable networking functions, and then, have those recast as web-class services. Meaning that they're not sort of a monolithic function that has a certain capacity, but they are more like a web service, which doesn't specify, or doesn't have a given limit. Rather, it grows elastically up and down, and makes use of various kinds of infrastructure, whether it's last generation, or next generation.

That is also pretty important in decoupling the services from the infrastructure so that you have a quicker time to market. And less, sort of, problems, or less work, when you want to upgrade infrastructure over time.

TT: What about web-scale's impact on microservices?

TA: I think that web-scale and microservices nearly go hand in hand. There are some exceptions, but, by and large, getting things to scale is really important. It's interesting, too, that a lot of the problem space is sometimes an inverse problem that you find in the big box data centers. They talk a lot about scale-out, but a lot of the telco opportunity is, in fact, scale-in. When I used to buy a box that was a, let's say, a packet gateway for mobility, it would come in a large size that would serve half a million customers, or something like that.

In order for me to make good use of a box like that, I'd have to put it in a fairly centralized place. With a disaggregated and microservice-oriented version of a packet gateway, I would be able to take very small, much smaller increments of capacity and distribute them much farther or deeper into the network, closer to customers, right? So I would be able to place functions and locations that previously didn't have enough aggregation to warrant that function that deep. That's a really exciting place for us because it speaks to performance, and a better experience for a lot of our customers. I don't know if it will happen or not, but it also looks like it could be a track toward intercepting applications that might need to be at the very edge of the network.

Related posts:

• AT&T's Anschutz Talks CORD & Containers
• AT&T's Anschutz: CORD Ties in Open Source, Containers

— Mike Robuck, Editor, Telco Transformation