You've designed some of the biggest networks in the world. Which one was the most challenging?
I would have to say the state of Alaska. There were so many outposts and remote sites [spread out] over such hostile geography and long distances. At that time, you had hundreds of mom-and-pop telephone companies, all in remote places, supporting a few hundred people. Putting together a structure was very challenging because you had such a wide range of options that you were forced to use. How do you deal with that?
When you're doing network architecture, you're one part psychiatrist and one part consultant, because you work with so many different groups that have different views. We have strategy sessions where people vote on particular choices. We'll present different approaches to a problem and the pros and cons and get their input. Then we do 'dot distribution.' We give everyone a certain number of dots and they are allowed to place them up on the board under the approaches they want to vote for. It creates a natural ranking. What has been the most interesting network to design?
Another fascinating project was an architecture I did for Hughes Aerospace. That project pulled all the major things that were going on in the industry together. We made choices on protocol standards for users, we defined an architecture for videoconferencing, we mapped a LAN architecture before LAN switching was really even popular, and we developed a WAN structure that tied all the facilities across the U.S. together.
We did that in concert with the business managers, to make sure that the requirements for the business were being met. It was when defense spending was being cut significantly, and Hughes had to move toward commercial markets as well. The time-to-markets are so different for the two segments that Hughes needed an infrastructure that would allow them to do concurrent engineering across multiple groups and consolidate a lot of operations to step up the time-to-market. It was interesting to help Hughes as they were going through a massive structural change and figure out how they could use IT to accelerate that change and allow them to shift with the market. What's the most common roadblock you see when you're designing a network?
The biggest roadblocks always seem to be human. When you're planning ahead, how do you achieve a balance between creating a design that serves you now and one that will easily adapt to technology that may crop up down the road?
You look for stability points. These are parts of the network that don't change over long periods of time. The best example is probably Ethernet. If you decided in the mid-1990s to standardize on Ethernet in the LAN, then you would have had a very good stability point that allowed for lots of different change in that technology. The same thing applies for the protocol stack around IP.
Another tactic is to base your management around SNMP [Simple Network Management Protocol], versus a whole bunch of proprietary agents scattered across your enterprise. In building new architectures, you've got to pick a stability point that you're going to bet will be around for a long time. And that is not a science; it's mostly art. In the IP telephony space, while SIP [Session Initiation Protocol] isn't totally finished, in the long term, that's what most of the client protocol is going to be. 802.11i could be another stability point in the wireless space. What is the biggest mistake that enterprises make when they're planning their architecture?
Not planning. It's like remodeling your house without an architect. You can do it by just working things out with the builder, but nine times out of 10, you don't get exactly what you wanted. A lot of little things don't happen if they aren't planned for, and a lot of people miss that step. What technologies will change enterprise networks the most in the next few years?
First, wireless switches can fundamentally change how LAN structures are being built. In LANs today, you have a three-tier structure: access, aggregation, and the core. Wireless is combining with a feature shift in which a lot more of the intelligence and the services are getting closer to users. This combination is moving toward an elimination of the aggregation part of the three-tier structure. So we're moving toward a two-tier LAN structure where you have access -- and lots of different kinds of access -- and those tie into a very low-cost and high-speed transport in the core.
Convergence is another area. It's been a buzzword, but now it's actually happening. IP telephony sales in small and medium-sized business are causing a structural change in WANs. Plus, Microsoft is including voice over IP in CE and the next version of Windows. We're in the very early phase now of really understanding what a whole converged architecture means. We've been talking about it from the point of view of low cost of acquisition and operations and facilities. Now we're starting to get glimpses of how to get the real productivity advantages of it. Having Microsoft include it in their productivity tools convinces me that this is an area that will drive growth in our industry.
The last area is network security. Security has been in vogue, unfortunately, since September 11. We had a knee-jerk response to that, installing lots of appliances all across our companies. We're now at the cusp of network security moving from point products and appliances to becoming a major feature set in basic infrastructure products.
FOR MORE INFORMATION:
Read the latest Lippis report on the changing structure of the enterprise WAN
Browse our Topics on network design