FreeWire

Which Will Collapse First, The Net Or The PSTN?

About 100 years of fine-tuning have made the PSTN the largest, most complex and most efficient machine ever built--for handling three-minute holding-time calls. Central office switches--those multimillion dollar, application-specific mainframes--and the trunks that connect them have been so cleverly optimized around a particular traffic model, they are at risk of becoming obsolete faster than the half-life of a single procurement cycle. The culprit? Twenty-minute holding-time Internet calls-- the tip of a data iceberg that will break the PSTN in two should flat-rate ISDN ever take hold.

Sobe ring Statistics "Don't be silly," you say. "My teenager talks on the phone for hours, and this hasn't caused Armageddon. You must be a shill for those greedy telephone monopolists that are trying to scare state public utility commissions into hosing hapless consumers." Greedy monopolists they may be, long overdue for a competitive comeuppance, but the problem is nonetheless real. And the lack of market-based pricing mechanisms that can attract resources to traffic hot spots can only make matters worse.

The problem is not hard to grasp for anyone who has taken an elementary course in probability and statistics. Voice calls follow an exponential holding-time distribution, with a mean of about three minutes. While 20-minute and longer holding-time calls certainly occur, their impact, when weighed against the sea of shorter calls, is vanishingly small. This distribution has held remarkably constant over the past century, with businesses showing a slightly higher frequency of call generation and differen t busy hours than residential customers.

Data calls to Internet service providers (ISPs), on the other hand, follow a power law distribution, with a mean holding time of approximately 20 minutes. In addition to the longer average duration, power law distributions have much longer tails. Because of this, the aggregate impact of these tails is much higher on total call-minute loads. Calls longer than 100 minutes are not a bit unusual, with nailed-up circuits becoming more and more common as people use ISDN to bypass outrageous leased-line rates.

As in any telecommunications system, traffic bottlenecks can occur in many places--the line card backplane, the switching fabric itself or the interoffice trunks. In some cases, call-processing elements can introduce bottlenecks, particularly if they were designed to do one thing and then are asked to do another (more on this later). The task of systems engineer ing is to balance the resources required to smooth out the bottlenecks while providing a consistent a nd predictable grade of service. In the case of the PSTN, a blocking probability of less than 1 percent is the norm. Ponder that level of performance the next time it takes you 50 call attempts to connect with your ISP's undersized modem pool. The operations and management systems required to track traffic growth and add network resources before troubles arise are massive things indeed, all tuned to the antiquated tariff structures of a regulated monopoly. (Compare this to your local ISP. "Hey Harry, do you think we should put in some more modems? The customers are starting to gripe.")

As long as Internet penetration remained low, the 20-minute-average-holding-time, power-law-distributed data traffic was dwarfed by conventional voice traffic so the LECs could cope with the additional load using their standard textbook procedures. But a funny thing is happening now that Internet penetration is crossing into double digits. These procedures don't work any more because the assumptions they are based on are no longer valid.

The first central offices to fall victim were the ones that terminate traffic to ISP's local points of presence (POPs), funneling data calls from an entire metro calling area into one spot. Then the interoffice trunks started clogging up. Line concentration ratios--the ratio between the number of subscriber lines and inter-office trunks--started getting wacky. Finally, switches themselves started to block at rates two and three times the design points. In some central offices, fully 8 percent of the total call-minutes handled by the switch are for calls with holding times longer than 24 hours! This is egregious switch abuse, akin to hogging the CPU of an old IBM 360 dedicated to payroll processing to run a screen saver with some nifty 3-D graphics. All of this is happening, by the way, in a yet-unfrozen regulatory environment in which the LEC earns no compensating revenues to pay for the addit ional resources required because most data calls are local, flat-rate calls.

Champion of t he Geek As if things weren't bad enough, onto this powder keg jumps Ralph Nader waving a blowtorch, demanding equal rights for modem users. Among the dozens of activist organizations he funds or promotes, alongside Ozone Action and the Council for Responsible Genetics (see www.essential.org), is the Consumer Project for Technology (CPT) (see www.essential.org/cpt/). CPT has been waging a highly vocal campaign to bully state public utility commissions into mandating low-cost, flat-rate ISDN. Ignoring the fundamentals of traffic engineering, CPT argues that a 20-minute local data call costs the phone company the same as a 20-minute teenage yak-fest, hence the price should be the same--free. It's as if a coroner examining a dead body with 10,000 pins sticking in it pulled the pins out one at a time, saying "Gee, what's the problem here? This one couldn't have killed him."

I tried explaining this to the dunderhead social engineers that populate CPT's ISDN mailing list some months ago, but I was kicked o ff the list for criticizing Nader's political motives for launching this crusade. The whole issue has become religious--part of the campaign to empower the "information have-nots"--and it has attracted major players, including Intel and Compaq, which should know better but are apparently willing to trash the phone network as long as they can promote more bandwidth-hungry applications that force people to upgrade to Pentiums.

Of course, the monopoly LECs are reacting in the only way they know how--running to the protection of the regulators, warning that if things continue going downhill, grandma won't be able to call 911 when she falls down and can't get up. What they really should be doing is beating the snot out of their switch vendors to develop ISDN line cards that can route data traffic onto a parallel ATM network before it ever gets to the circuit switch. In fact, packet-mode ISDN, part of the original ISDN specification, is the most attractive approach for solving this problem, but this service ha s never been made available to the general public. Part of the reason, aside from the fact that there hasn't been any demand up until now, is that the call-processing hardware in the central office that normally handles the leisurely tasks of call setup is just not up to the job of handling high-speed packet switching. The other problem is that once a proper architecture is put in place to handle data, the LEC might just as well become the ISP, since there will be little added value for a third party, whose biggest contribution was gaming the tariffs. As you can imagine, a LEC takeover is hardly a popular solution in the ISP community.

What About xDSL? The various flavors of Digital Subscriber Line (DSL) that are floating around address the switch and trunk congestion problem from the get-go, separating circuit-switched voice from packet-switched data as soon as the subscriber line enters the central office. But we are years away from seeing widespread deployment of these emerging technologies. ISDN will be the only mass-market option for breaking the 28.8-Kbps barrier for at least the next five years. In addition, no other solution besides ISDN has such broad applicability to the wide variation of subscriber loop lengths and conditions found in the real world. A concerted effort on the part of the LECs, and some timely innovation by the switch vendors that serve them, could actually give crybabies like Nader's Raiders, along with the rest of us, exactly what we want--higher-speed, flat-rate Internet access.

If not, we may prove Bob Metcalfe wrong the hard way. At the rate things are going, the on-ramps will all collapse before the Information Super-Highway itself gets congested, sparing us the humiliation of nationwide data gridlock. Oh, what an ignoble fate.

Bill Frezza is the President of Wireless Computing Associates. He can be reached via e-mail at frezza@interramp.com.