AT&T is pouring billions into its network to make it more dynamic, which is resulting in new capabilities for enterprise customers. Network World Editor in Chief John Dix recently stopped by AT&T headquarters in Dallas to talk to Josh Goodell, VP of Network on Demand, about what the company is learning from early adopters of its Switched Ethernet on Demand service and what comes next. Among other things, Goodell explains how provisioning now takes days vs. weeks, service profiles can be changed in seconds, and how he expects large shops to use APIs to connect their network management systems directly to AT&T controls. Oh, and a slew of virtual functions are on the horizon that will enable you to ditch all those appliances you’ve been accumulating.
Let’s start with the big picture view of AT&T’s dynamic network efforts. What’s the goal?
Usually when I talk about our strategy I start at the network access layer. This is the physical infrastructure that AT&T has built over years – the fiber network and technologies like LTE on the wireless side, and what we call Lightspeed, which is a combination of fiber and copper. It’s a very robust network that has a tremendous reach and tremendous speed. All of that is foundational to what we’re doing now. Our Network on Demand platform acts like rapid onramps to that very fast network. So that physical layer is important and one area of the overall puzzle.
Another area is driven by John Donovan, Senior Executive Vice President of Technology and Operations, who is driving our software-centric architecture. We’ve called it different things over the last couple of years, including Domain 2.0, but at its core it’s about driving virtualization within our own network. He’s made the commitment that by 2020 we’re going to virtualize 75% percent of our network. That’s all about driving up utilization in the network and enabling scale and flexibility.
The third piece is enabling these same types of capabilities for our business customers, and that’s really where Network on Demand comes into play. It’s taking technologies that we’re utilizing internally and making our core strategic services better by utilizing the same technologies.
The Network on Demand platform initially launched with one capability -- AT&T Switched Ethernet on Demand -- and the second service that will launch is Managed Internet Service on Demand. Then we will continue to add additional services over time. So Network on Demand is creating this platform that enables customers to have a rapid onramp to that very robust network.
That gives customers more control of their network, the ability to rapidly scale up or scale down their network, and improves TCO, not just because you have the ability to use exactly what you want, but also because you can be more productive. You can spin up a location more rapidly than you could have in the past.
Then we will also start getting into services that take advantage of both SDN and NFV, where you’re actually virtualizing what has typically been purpose-built appliances. We don’t have a product in the market yet but we’ve announced that the first iteration will be available in the next few months.
We’ve reorganized our entire technology organization around network simplification and a software-centered network, and then exposing those capabilities to our customers. That’s the big picture and Network on Demand is one piece of that picture.
It’s important to understand that we have a lot of conviction across all three of those areas. From 2009 to 2014 we spent about $140 billion in those three areas. These aren’t hobbies. These are how we’re committed to drive a differentiated network experience.
How long has the switched Ethernet service been available?
We opened the first market in Austin Texas in November 2014, expanded to five markets in February of this year, and in April expanded to 170+ markets. So that is very, very fast for any service we’ve ever stood up. Part of it is the technology. It’s different. It’s building on a software layer that allows for a rapid product instantiation, but we also used an agile approach to development, a DevOps model, and the combination of those things allowed us to move at a rapid pace.
When does the Internet on Demand service go live?
Managed Internet on Demand is CI in Atlanta. Interestingly, with AT&T Switched Ethernet on Demand there’s no virtualization happening. It is an SDN layer on top of existing network infrastructure. Managed Internet on Demand is a different architecture that actually takes advantage of both SDN and NFV, so we’ll be virtualizing the customer edge. Typically the customer has a router on-premise that will be virtualized in the AT&T cloud, and then we will also be virtualizing the provider edge. It’s a big deal because I expect that over time we’ll be virtualizing a lot of different services, so we’re going through what it takes to do this with this next service for the first time ever.
Does the customer have to virtualize the router?
The initial offering will be a virtualized router only. If they want they can buy a switch and put it at the end of their network and run something off of that. Eventually we’ll have a use case where we’ll actually deploy a piece of CPE on-premise they can use, but the initial use case is a pure virtualized router.
Coming back to the Ethernet On Demand service, how many customers do you have?
It’s been really interesting to see the way that has played out. As of today it’s over 350 customer networks, about 1,000 locations. That’s a lot more than what we had expected. Market demand has been pretty strong.
Is there a typical customer profile emerging?
It has been across industries. The largest network we have provisioned is about 150 locations and as small as two-locations. So it’s run the gamut. It is more prevalent so far in the mid-market and down-market, but every single one of our segments has seen traction.
One interesting thing is how we’ve simplified the selling experience. We’ve enabled our sales people to use an iPad to order the service and do all of the contract work with the customer on their premise. Historically the presale cycle alone was days. Now everything can be done in one sit-down discussion. That’s not the cool, interesting technology that SDN represents, but it is an interesting case of how, when you take friction out of the experience on both the seller’s side and the customer’s side, you’re going to see traction and we’ve seen it with Ethernet.
How long does it take to deliver Switched Ethernet On Demand?
When fiber is available to the customer building, and if you take out the “Customer not ready” situations, it’s five days. The equivalent when you’re in a fiber location but not on Network on Demand is probably closer to four weeks. When you don’t have fiber availability the cycle time obviously goes up because you have the build process, but we’re still automating the overall process with Network on Demand.
How much are early customers actually changing the Switched Ethernet service profile once they are online?
There are a couple of things customers can do, one of which is to add locations. Customers can go into the portal, which knows the inventory of their locations, and go from, say, a two-location network to a three-location or a four-location network in a matter of days without ever having to talk to anyone. That’s a big deal. Historically that would have been multiple phone calls and a fairly long provisioning cycle. Customers can now do it themselves at their own convenience.
And another common use case is the ability for a customer to scale up or scale down their network. Any business that has seasonality is going to be interested in that use case. For example, we have K-12 as well as college institutions that are very interested in that capability.
We’ve also talked to a few hospitals that have branch locations that do analysis on medical images that are interested in the ability to scale up the network to send large payloads and then scale it down again, which isn’t something they could do before.
One of our large customers uses Network on Demand service for redundancy between data centers. They have this as a secondary network and keep it scaled all the way down, and in the event they have an issue they can scale it up within seconds.
Another interesting use case is around rapid provisioning for M&A activity. After a merger or acquisition the network may go from 10 locations to 20 locations literally within a week, and provisioning agility is important for those types of situations.
Does it surprise you that the mid- to smaller-size shops are the early adopters? I would think the largest shops would be dying for these capabilities?
It does surprise me a little bit. I think there are a couple of things at work. I mentioned that customers manage their networks through a portal. Some of our very large customers will want to have their own network management tools tap directly into our network through APIs. We expect to do that. We just haven’t gotten there yet. In fact, I expect at some point we will federate our SDN Layer 2 network with other carriers that have SDN Layer 2 networks. It’s a very natural evolution. And federating the network with a large customer’s network management tools is just another version of that. It’s more of a northbound API as opposed to an east-west interface.
I also think we have a service now that hunts very effectively against competitors that have been attacking us down market.
Going back to bandwidth flexibility, obviously the seasonal use cases make sense, but are other customers tweaking the settings more or less often than you would have expected?
Less often than I expected. When we started we limited the number of changes allowed to one per day because we had no idea what the actual behavior would be. What we’ve seen is customers aren’t going in and ratcheting it up and down frequently during the month. They may do it once or twice, but it’s not as prevalent as I would have expected. It’s still early, though. Since we’ve really been at scale only since April, it’s a bit early to say how much the behavior is going to shift.
What are the increments you can scale up and down?
It goes all the way from 2mbps to 10Gbps with several increments along the way.
You said you’re in 170 markets. How many states is that?
Our incumbent AT&T 21-state footprint. We do have fiber assets in other areas, including New York, Philadelphia and Boston. Those markets are not available yet on Network on Demand, but we expect that we will bring them on net in the future. For now the Ethernet on Demand service is limited to our 21-state footprint and the 170+ markets.
Internet on Demand is up next. What comes after that?
The next capability after Managed Internet is what we call Network Functions on Demand. Network Functions on Demand is basically re-looking at how premise-based appliances are used. Typically today you have purpose-built appliances, whether that’s a router or a firewall, a WAN accelerator, you name it. In the future we will offer what we call universal CPE that can run multiple virtual functions, so software instances of those capabilities on that universal CPE platform.
We are also building the ability to deliver those same types of virtual functions directly through our AT&T cloud. So you can envision a time where customers will have a series of capabilities that are delivered both through a universal CPE on their premise -- again, things like router functionality, firewalls and WAN accelerators -- as well as capabilities delivered directly through our cloud.
There would be advantages to using one versus the other. For example, an application that is going to be shared across multiple locations, you probably want to use more of a cloud approach, whereas an application that’s more specific to a location will sit on the universal CPE.
That capability will begin to roll out in the next few months. It will evolve dramatically over time. The first instantiation of it will be that universal CPE capability with a virtual router. Then we will add other virtual instances within the portfolio over time, and I expect it will evolve pretty dramatically throughout 2016.
I presume the universal CPE is a server that you manage?
Yeah, it’s a white box x86 server modeled to run three to four virtual functions. It’s got a Gig of throughput, so it’s a fairly robust platform. The box will be managed by AT&T. The virtual router will be managed by AT&T. But I expect over time you’re going to have multiple virtual functions on this box and we will have options for both AT&T managed as well as customer managed functions.
Will the virtual functions be available from different suppliers?
Yes, it’s an open platform and an ecosystem of partners. We’ve announced some partners and the ecosystem will expand over time. We’ve announced Juniper, Cisco, Brocade.
The appeal to the customer is fewer appliances to manage?
There are different value propositions with the universal CPE concept. One is you go from having multiple boxes to having one box. That’s a big deal. Just from power consumption and having less to worry about, that’s a big deal.
The other thing that is important is, because the functionality is being delivered through software that can be downloaded at any time, the issue around box obsolescence is less of a problem over time. And the installation cycle time agility plays out here as well. Historically, if we were to install multiple boxes on a customer premise, that typically happens sequentially and it may take 30 days for the first one and upwards of 90-120 days all told. In the future this is a plug-and-play model.
Does NetBond fit into this picture?
As we talk about the AT&T SDN story, NetBond is an element of that story. NetBond is basically secure connectivity to third-party clouds. Today if a customer wants to take advantage of NetBond and AT&T’s Switched Ethernet on Demand, they can. In the future I expect they will become more and more integrated and it will just be an extension of an overall on-demand experience. They both can be used today but they’re not fully integrated through one management pane of glass.