Though IP is now an easy sell, the question is how to make it reliable enough to handle increasingly complex networking needs.

As director of strategic marketing and technologies for Nortel Networks' enterprise solutions group, Rybczynski's latest work involves financial institutions trying to keep their networks modernized and IP-aware. "What I really enjoy is translating all this stuff for a customer. Why should you be excited about this? Is it because you read about it with all these dot-com companies?"

Rybczynski admits that 28 years in one company is unusual these days. "Nortel always offered lots of different opportunities. So rather than change jobs, change cities, change countries-I opted to stay."

IP was a relatively radical idea when he started. "Much the way we evangelize around policy management or optical networking and how it will change the world, we had to evangelize packet networking," Rybczynski said. "When I started, the Arpanet-the predecessor to the Internet-was running on minicomputers, slow speeds, many kilobits [per second]. We came along and said this new stuff called packet switching is the wave of the future."

Nortel had actually proposed IP as a means for packet-based switching. But the industry instead turned to a more familiar, circuit-based approach-one that has clearly defined end connections, as with ATM or frame relay.

"The reason was, you could control it better," Rybczynski said. "It obviously gave you higher security, was easier to bill for and was more familiar in [terms of] how to control."

IP was perceived as difficult because it is connectionless; that is, anybody can connect to anybody else at any time. That's still troublesome today. Telephone calls, for example, rely on permanent connections between points, which complicates services such as IP telephony. But ironically, it's IP's connectionless nature that has helped it flourish.

"The real value of the Internet and these IP networks was that you could connect to anybody. It was a really democratic kind of environment," Rybczynski said.

A few large customers actually did understand the benefits of IP. "One of the first two customers was the Federal Reserve Board. We ran the economy for the United States," said Rybczynski.

In modern networks, which are being groomed to handle data and voice traffic simultaneously, the spread of IP creates a need for ways to handle quality-of-service (QoS), which would give higher or lower priority to various types of traffic.

Telephone calls made over the Internet would need to stay coherent on the other end, for example, while e-mail can afford to be queued up for a few more seconds.

"It's very difficult to predict how much traffic you're going to have and where the traffic is going, and likewise it's difficult to predict where you're going to have failure points," Rybczynski said. "Another way of spinning that is that reliability is very, very important. But reliability these days is as much dependent on how you manage the traffic-for instance, on failure conditions or peak load conditions." It's those aberrations that will require QoS.

Policy-based networking has two components: what goes into the router or switch itself-such as hardware queues, software algorithms for priority, etc.-and the policy manager. The policy manager "doesn't operate on the individual bits of traffic flying by. It's sort of like a server," Rybczynski said.

The policy manager communicates with the different switch elements through a protocol such as Common Open Policy Services (Cops), which Nortel and others have espoused. One difficulty lies in getting QoS implemented without slowing down the router or switch's operations, since typically, any additional application will drag down speed. Ultimately, "it'll be built into the hardware."

One particular area of interest is to allow applications themselves to request bandwidth or certain performance levels. Applications would set the rules-within reason-rather than the network itself doing so. Cops could be used along those lines, Rybczynski said. "The policy manager would be involved in real-time."

Enterprises in general are "taking all these changes that are happening on the Internet and optimizing them for internal use," Rybczynski said. "With the Internet and the 'revolution,' if you like, that's been taking place, there is a bigger opportunity for enterprises in how they conduct business internally."

IP will permeate all networked applications, Rybczynski believes, but it won't take over the entire network. In connecting branch offices, for example, banks often prefer to go with "wavelengths"-that is, they use a slice of dedicated bandwidth on a fiber-optic line. "When you're connecting two major sites together, even if everything is IP-based, the most effective way is just to gimme the damn pipe," he said.