Hossein Eslambolchi is a man of many titles. He is president of AT&T Global Networking Technology Services and AT&T Labs, as well as CIO and CTO of AT&T proper. When Hossein talks, I listen. And what he talks about is the inevitable move to 100 percent IP networking.

"IP will eat everything," Eslambolchi tells me.

Driving this convergence is the ascendancy of MPLS (multiprotocol label switching), a router protocol that, among other things, recognises the kind of packets it is sending. With MPLS's capability to slice and dice services in as many ways as your imagination will take them, we will soon be flooded with options for managing and automating services at far lower cost.

Given that over time all disparate applications will move to IP - including voice, video, and data - Eslambolchi thinks it prudent to talk about some of the more common mistakes made when designing such converged networks.

Many companies think that if they have a subscriber base of one million, they will be able to easily "scale it in mid-air" as the number of subscribers increases beyond that point. Eslambolchi says what these companies don't realise is that as subscriber count goes toward that second million, the complexity increases by a factor of 100.

"You need to worry about operations, maintenance, and scale while you're still on the ground. Otherwise it is like trying to change engines in mid-flight," he adds.

Too many experts believe IP has an infinite amount of capacity. Try watching the space shuttle on Yahoo alongside a million other subscribers and you will see that this is absolutely untrue. Networks must be designed for peak.

Poor standardisation is always a culprit when errors occur. Proprietary software in IP is prone to many more defects, Eslambolchi says.

Also, he says, network engineers should beware of "unbridled complexity" when writing code. IT must write code with an IP mentality or it will come back to haunt the network. This includes keeping it simple and building in reliability and security from the beginning.

Although it will be difficult to adhere to this one, Eslambolchi warns that an unwillingness to move from legacy services to IP will cause headaches down the road. "By trying to operate two different environments, your TCO will be significantly higher than putting all of them on IP MPLS," he says.

But let me leave you with Eslambolchi's number one IT error: failure to push more security upstream into the cloud. It is a big mistake to put all security at the edge of the network, he warns.

"In the 90s the IT bigots said, 'We are going to make the network dumb and end points intelligent,' " he says, adding that it is nearly impossible to scale authentication and encryption to every end point. Which would you rather have, he asks, a thousand places for protection or protection in one location?

Eslambolchi's point is that, although you need protection at the network edge, there are limitations. So get off the bandwagon that everything goes to the edge. It will be the combination of edge and centralised services that empowers IP and gets us out of the mess brought on by hacking, worms, and viruses, Eslambolchi says.

Yes, letting someone else manage the cloud is going to cost. But if you don't, you'll more than likely end up paying in another way.