Fault lines

Two years ago, an executive with a competitive fiber operator bragged to me about his company's futuristic IP-based network. Intrigued but skeptical, I questioned the reliability of an IP-based telecom network. After serious prodding, he admitted the company also operated an ATM network - for backup situations - and could guarantee its IP services because it used only 4% of its bandwidth. Because so little traffic ran across the network, conflicts rarely, if ever, occurred.

The thought has haunted me since. Such low network utilization might work in the short term, but that model won't sustain as the bandwidth fills up. Service reliability is affected as usage on a best-effort IP network increases.

And that brings up an interesting question: How reliable does the network really have to be? Telecom executives maintain that no data can be lost. Lifeline services translate to five 9s reliability. In contrast, datacom executives argue that a lost packet here or there won't matter. It will be retransmitted and recompiled at the destination point. The debate rages; allegiances are based on cultural perspective.

The prevalence of IP is pushing the reliability argument from an academic one to an engineering one. IP is the protocol of the future. Service providers tout "IP over X" solutions, and the goal of products such as Lucent's LambdaRouter is IP over optics. Application developers concentrate on creating IP-based services. Traditional telcos are bolstering their sagging voice revenues with high-margin data services. Data CLECs - already in that game - are deploying services that make the network more critical to users, thus upping usage and, in turn, revenues.

But as IP takes hold, its fallibility emerges. Hailed for its ubiquity, it's equally railed for unreliability. Poor QOS, the lack of service level guarantees and an inadequate addressing scheme are common complaints. Some criticisms are valid: IP's addressing mechanism is reaching its limit much more quickly than expected. The preponderance of cell phones, pagers and PDAs is stretching IP to the limit - and that's before you add the home of the future with networked air conditioners.

The other concerns are less cut-and-dry. Cisco, probably the biggest proponent of IP acceptance, insists that IP can be as reliable as traditional telecom networks. IP with guaranteed service levels? Sounds ridiculous - until you hear the rest of Cisco's point. With proper network engineering and bandwidth management techniques, IP can maintain five 9s reliability.

Perhaps that's true, but those techniques introduce unnecessary complexity - and the people familiar with those techniques cost a pretty penny. In a time when all technology companies are desperate for talent, Cisco's solution seems to come up short. Networking should be getting easier, not harder.

Some providers are willing to use such workarounds, betting that the end result - an advanced IP network - will be worth it in the long run.

In my mind the deeper issue is the definition of reliability. Is a lost packet a failure on the part of the network? Does the retransmission of data indicate the network is unreliable? Simply put, no. If the information arrives as it was intended with little or no delay, the network has succeeded in transporting that data. For most applications, that's adequate.

Die-hard telecom execs claim that customers won't stand for poor service quality. I disagree. Cell phones drop calls, and we can't live without them.

Service providers that embrace IP are making a move for the future. The world may not be ready for lifeline IP telephony, but neither are traditional telecom carriers in danger of extinction.