AT&T Broadband made a lot of news last year. Unfortunately, much of it was not the kind of news shareholders wanted to read. But on the flip side, AT&T CEO Michael Armstrong tapped three key "cable guys" to step in and run the Broadband business unit in an attempt to restore order and improve operating margins.

One of those who rode in wearing a white hat was David Fellows, who showed his operating prowess during his years at Continental Cablevision and MediaOne. Although Comcast Corp. is preparing to take over AT&T Broadband (perhaps later this year), Fellows is making AT&T's technology decisions in the meantime.

We caught up with David to get his latest thinking on a wide variety of subjects related to advanced service rollouts. What follows is an edited transcript.

What are your technology priorities for the rest of this year in video, voice and data?

The one big thing this year is upgrade. Getting the rebuilds and upgrades restarted, and then just upgrading our plant. We announced 62 percent was upgraded, and we will exit the year at 70 percent. That's the overarching, overall priority for the company.

Within each of our businesses there's a "product" thing, and then there's an "infrastructure" thing. Within high-speed data, (the focus is on Broadband Choice, the multiple ISP product). But there are also a number of things beyond just multiple ISPs. It also includes things such as measuring consumption, dealing with various tiers that the product management group might want to launch, dial roaming and remote access. That's one big thing.

David Fellows ‘Retail is the law of the land, and you’d love to have some people with retail presence and consumer brands out selling consumer devices that can hook onto your network in a secure manner’
—David Fellows

The second thing in high-speed data is, we built a network and moved our former Excite@Home customers on to that. We also have a number of former MediaOne properties. It's now taking those two networks and making them look the same, merging them together, and then toward the end of next year taking control of more aspects of that network.

In the telephony space, the product piece is advanced intelligent network features. A number of features that are achieved not through software in any particular class 5 switch, but through an external processor, a feature processor, that controls the switches. We are doing that AIN service in such a manner that the feature processors can control not only the Nortel and Lucent switches that we have on our circuit-switched side, but the same features could be rolled out onto voice-over-IP softswitches.

The second priority in the telephony group is voice-over-IP. This year (we'll be) doing lab work and product integration, and then taking that to the field next year.

In the video space, we have video-on-demand (VOD) rolled out in a couple of regions. One of the things that I'm doing there is adapting the architecture of VOD so that there's an Internet protocol infrastructure–a metro area network between generic servers, and then edge devices that pull the Internet protocol MPEG signals off and put them out into the neighborhood.

We're also continuing to work in the PVR space. We have a business arrangement with TiVo in a couple of our markets, where we have a TiVo box sitting next to one of our Motorola set-tops, and they communicate back and forth. A third thing that will come toward the end of this year will be high definition TV.

Let's follow up with multiple ISP tests. You did a test in Boulder and you have one coming in Boston. You've also done some deals with regional ISPs. What have you learned so far?

Coming out of Boulder, we tested a number of things. One was at the plumbing layer–how to actually take a piece of packet and route it to the ISP of choice. The basic way we do that is with policy-based routing. Even though that's the way that we're doing it, we continue to look at a number of tunneling techniques and technologies up through perhaps the most sophisticated being MPLS, but right now, policy-based routing looks like the best way to go.

There's also the provisioning layer, also known as the B-to-B interface. I think back to the early days in Rochester, N.Y. of Time Warner trying to get in the telephone business, and all of these faxes that were passed back and forth between New York Telephone and Time Warner. These are the things we want to avoid.

Part of that is also monitoring the networks. If there's a slowdown, if the customer calls in with a problem, how do you handle it? How much of that can be automated in terms of visibility into each other's network?

How difficult are you finding it to meld together the back office for Earthlink and for yourself? Is that where the biggest problem is–getting it all coordinated?

That's what the biggest piece of the Boulder trial was. We spent tens of millions of dollars, and Earthlink is quite familiar with that. They know what they have to do on their side of the interface, and we know what we have to do on our side. These new, smaller players that we're signing up, I expect that we will have to supply support to them to nurture them, bring them onto our networks, teach them how to do this, and it's part of what we're signing up for with Broadband Choice. So yes, it is a huge piece of work.

The fact that the number of (alternative) ISPs is measured on one hand is not due to technical limitations. We are out trying to garner interest and sign up any ISP that's interested in dealing with us. There are a lot of process issues. For example, in New England, that is one of the former Roadrunner markets that I need to get on our own backbone and provisioning system. But we know what to do; we know what the answers are. It just takes longer than we hoped to get there.

What about the transition to DOCSIS 1.1? What's your thinking there?

There are two aspects to that. One is that our installed base is largely Cisco CMTSs. I have expressed my desire to move that installed base to 1.1, and we're in the process of planning that. It will take longer than the rest of this year to bring that base onto 1.1, but we're working with Cisco on proposals to do that.

1.1 has also, in a sense, been coupled with carrier-class, ready-to-rollout PacketCable, and so in a separate program, we are selecting what we believe to be carrier-class CMTSs that run DOCSIS 1.1, and will be the basis of rolling out our voice-over-IP initiative.

There seem to be some people who think we have to take the Cisco equipment and throw it away and bring in new carrier-class equipment, but that's not my plan. (I plan) to launch DOCSIS carrier-class carriers next to DOCSIS 1.1 legacy carriers. Because I believe that there are also a lot of services that can be assisted by 1.1 that are not primary line telephony offerings. For example, gaming. You might be willing to pay extra for low latency and DOCSIS 1.1 has the tools to allow a low latency offering, but it's not lifeline.

Let's get into voice-over-IP. What technology are you testing related to gateways and soft switches?

We had working in the lab some time ago what is sometimes called an IPDT, which would be an IP carrier throu