When usage matters: Broadwing's Gigabit in the MAN

By Kevin Tolly
Network World, 08/06/01

With the likes of Yipes Communications and Cogent Communications already deploying high-speed Ethernet metropolitan-area networks, Broadwing Communications' May announcement that it will deliver Gigabit Ethernet interfaces to the customer premises did not come as a shock. What did take me aback was the statement by a Broadwing vice president that customers would pay only for what they use. To me, usage-based pricing and LANs are a deadly combination.

Let me begin at the end. Since the inception of LANs at the barely-a-megabit level, LAN designers have, essentially, ignored traffic. Given the relatively quick ramp up over the years from low-speed shared to high-speed switched LANs, it has rarely been worthwhile to spend the time to figure out what traffic was using up bandwidth that would otherwise be unused. The cost of that Fast Ethernet or Gigabit uplink was fixed, regardless of the usage. So long as no bottlenecks existed, no attention was needed.

That, of course, would all change were a customer to move into Broadwing's MAN. While specifics of pricing were not disclosed, the bottom line is the same - you'll pay "by the pound" for whatever crosses your link - whether you've asked for those bytes or not.

Imagine the accounting nightmare. A full-duplex, Gigabit Ethernet link running at full load, day and night, could push through (according to my back-of-the-napkin math) some 650 MILLION gigabytes (not bits) of traffic over the course of the month. What would your bill look like? Where would the breakpoints be?

More importantly, how would this change the customer's behavior?

Today, network managers are content to filter out spam and unwanted attachments at the edge of the network. With the usage-based model, the damage would have been done - economically speaking - before the e-mail arrived at the filtering point.

One can imagine new kinds of hacker attacks where these hackers keep a steady "drip, drip, drip" of, say, large "ping" frames going across the link but staying "under the radar" of the firewalls.

The enterprise-caching business promoted "saving bandwidth" by serving content locally. Overall, companies have greeted this technology with a yawn. With edge traffic shapers hooked up to links that are not usage sensitive, network managers don't seem really to care if the same 80M byte service pack gets downloaded by different (or the same) people time and again across the Internet.

I have to wonder whether Broadwing has thought this through. The approach that it is taking is fraught with problems.

Not only does Broadwing have to make sure that the usage tracking system is accurate - and can deal with these kind of large numbers - it has to make sure the accounting mechanism doesn't negatively affect system performance. Historically, "intensive" monitoring has exacted a toll in performance.

Furthermore, the company is going to have major headaches with resetting the counters at the end of an accounting period. And what happens if a system failure in a switch wipes out the utilization information just before the end of the period?

And, unlike water, gas or electric meters, the customer can measure his utilization on his own. I can almost guarantee that multi-Gigabyte "irregularities" will surface when customers compare the usage they've monitored with what Broadwing supplies.

And, assuming the myriad technical hurdles can be overcome, how on Earth is Broadwing going to change the way customers design their networks? This is just not going to happen.

Unless Broadwing wants to move deeply into the services space and provision e-mail servers and traffic shapers on their side of the network, pay-as-you-go might be something it wants to reconsider.