Goeller on Telecom Traffic

AOL Folks At Home

(Business Communications Review, May 1997)

Almost 40 years ago, during my Bell Labs days, a friend got stuck in a task force whose mission was to study the possibility of nation-wide flat-rate toll service. As of 1956-7, AT&T's nationwide microwave network was only 5 years old, crossbar switching for Direct Distance Dialing was being introduced at a great rate, and AMA, or automatic message accounting, was going in with it. But detailed billing, even when automatic, was expensive. Replacing it with flat rate would save an enormous amount in data processing, to say nothing of switching system complexity in those days preceding stored program control and its derivative technologies.

As I heard the story, nationwide flat-rate would have been possible for an increase of 50 cents a month, IF there were no increase in long distance calling. But that was a big IF, and AMA installation continued. What was later to be known as the "freeway" or "uncorking" effect was well understood by traffic people even in those days at the dawn of time: if you do something to improve traffic flow, traffic will increase almost at once to make it much worse than it was before.

Fifteen years later I got to see this effect first hand when RCA took over the toll network in Alaska from the Military. Civilian use of the military network had not been encouraged, and when RCA tried to expand civilian service to meet the needs of the rapidly growing Alaskan economy (the pipeline was just about to go in), everything they did encouraged so much more traffic than they anticipated that congestion got worse than it had been before. It took heroic measures to save the day.

What was ancient history in the telephone business is being repeated by America On Line today. AOL decided to double its previous monthly charge (to $19.95). For those who complained, AOL was going to point out that it now offered unlimited service rather than just 5 hours at $9.95 with additional hours at $2.95 each. Both the New York Times and the Wall Street Journal discovered and reported this proposal toward the end of November, somewhat spoiling the surprise.

Christine Gregorie, Attorney General of the state of Washington, actively intervened (NYT, 11/25/96, p D9). Ms. Gregorie insisted such surprises violated the state's consumer protection laws, and AOL agreed to notify its customers with an electronic message at log-on, to offer refunds for those who did not want to be upgraded, and to provide a "life-line service" at $4.95 for three hours a month.

Other than Ms. Gregorie, I am apparently the only person in the country who was incensed by AOL's furtive price rise. I took advantage of the el-cheapo rate which is all I need for e-mail. Everybody else seems to have been overjoyed with flat rate service. They logged on and have stayed logged on ever since. At least those who got their first. The laggards are getting busy signals—and complaining vigorously.

Other state regulators listened and, led by Jim Ryan, Attorney General of Illinois, convinced AOL to provide refunds for those most excluded (NYT Jan 30, 1997, p1). And now, AOL's 800 number, which you call to get a refund when you can't log on, is giving as many busies as the network itself. One can't help but wonder if Dilbert-land conferences are being held to choose between increasing the number of network ports or the number of 800 lines used to receive complaints about lack of network ports.

What the Problem Is, and Isn’t

Although the promised refunds and free time (if you can log on) may be helpful to the injured customers, the local telephone companies are left with a problem. Unfortunately, they don't seem to know what it is. In a Bell Atlantic report to the FCC quoted not only by our own Eric Krapf (BCR, January, 1997) but also by David Kopf in America's Network (Dec. 1, 1996), they claim the problem is the high occupancy of modem ports at the points of presence (POPs) which customers dial to access AOL and the other Internet Service Providers (ISPs). Cloaked in bizarre gibberish based on such obsolete traffic units as the CCS, they indicate that access port occupancies of 90% and higher are likely to cause the destruction of the telephone network.

That this is silly can be seen from elementary traffic theory (see Table 1). If high occupancy alone were the problem, any telephone company serving a metropolitan area with a few dozen medium sized PBXs would have been blown out of the water years ago. Because telcos have been able to serve PBXs for the last 75 years, their problem has to be something else.

To see what, consider a bigger PBX with 165 incoming trunks, like the one RCA had in Camden (which I once helped manage). With this large trunk group, occupancy is well above 90% at B.05 (Erlang B) service, all morning and all afternoon. But these were 5 minute calls. Almost 12 per hour per trunk, or better than 1900 calls total in a busy hour (around 12,000 calls a day). Now, suppose they were not voice calls, but people dialing AOL and planning to stay for an hour. The same 165 trunks could only handle 165 calls. What happens to the 1700+ calls that would be served if holding times were 5 minutes?

They retry over and over, getting madder and madder while making free use of DTMF detectors, busy tone trunks, local and tandem switching, etc., etc. This furious activity, tying the telephone system in knots by the people who DON'T get through, is the actual problem.

But traffic theory has a little more to tell us about the problem. As Table 1 shows, the bigger a trunk group is, the higher its occupancy per trunk can be for the same grade of service. Somewhat less well known is that the bigger the trunk group, the more sensitive it is to overloads (see Table 2). One minute you're doing fine, and the next, you get a surge of traffic, all trunks become busy, and the number of calls turned away escalates.

The New York Times (11/25/96) describes a central office in Santa Clara with 2000 trunks to an ISP. With the meter running, users might get off when finished, sharing the trunks with others and allowing everybody a shot at access. But with the new flat rate, where you have to stay connected to get your money's worth, we have a whole new ball game. What happens is people come home from work early, log on to "get there before the hoarders," and stay logged on all night. The ISPs have developed software to disconnect customers who don't send or receive anything for five minutes, and the customers have responded with software to create just enough activity to fool the ISP software and hold the connection. With voice traffic, it is well known that when people have trouble getting through, their holding times are longer when they do make it. The same thing is happening here when you know you can't get back on if you end your session.

Remembering What Packet Switching is All About

Clearly, the problem is not the high occupancy of each ISP trunk, but the long holding time of each ISP call, greatly reducing the number of customers each trunk can serve in the course of an evening. Although AOL was unable to predict this behavior, the telcos have a long memory; in New York City, back in the 1960s when time-sharing was the state-of-the-computer-art, uptown stock sales offices would dial up a connection to their downtown main-frame computers, getting a tie-trunk connection by making a daily local call.

One solution was to speed the elimination of flat rate business lines throughout the country, and also to make message-unit calls time dependent, clicking the message register every five minutes instead of once per call. That probably encouraged the use of real tie trunks, rather than "permanent" dial-up connections with local and tandem switches inserted into them.

Eliminating flat-rate residential lines today would cause a bit of a stir, particularly when critics are pointing out all the money telcos are making by installing "extra lines" for Internet users, and AOL and its fellow ISPs are installing additional access lines as fast as the telcos can supply them. But unfortunately, extra lines, either at the customer or the ISP, are not the solution. Again, a basic bit of fundamental information is being ignored.

Once a customer gets past the ISP's modem, he or she has left the land of CIRCUIT switching and entered the world of PACKET switching. And this is the fundamental point: the purpose of packet switching is to NOT occupy switching and transmission facilities when you are NOT sending anything. That way, the same facilities can serve a very large number of customers who are sitting and thinking, reading a previously downloaded file, or eating supper.

The big ISPs have their own packet networks to gather up traffic from several POPs in each LATA and connect it to their centralized servers several hundred or even a thousand miles away. Access to the Internet is through these servers. The ISP networks use broadband channels on optical fiber, and use them very efficiently by only sending packets when there is something to send.

The bottle neck is logging on. The user has to dial up a circuit-switched connection to the POP, via one or more local telephone switches, a connection that is in use for the whole duration of the call, whether or not the user is sending or receiving information. Similarly, the ISP's modem that the customer (sometimes) reaches after dialing the ISP's number is tied up for the duration. These modems, one for each logged on customer, all feed into some sort of computer at the POP which, like the TIP (Terminal Interface Processor) of the old ARPAnet, convert between customer asynchronous data and packets between the POP and the server.

Extend Packet Switching Down to the User

Putting in more access lines and modems will give more users simultaneous access to the ISP, but that approach is expensive for the ISP and won't solve the telephone company's problem. Further, from the reaction we've seen so far, it is likely to expand the Freeway Effect. What is needed is to extend packet switching all the way to the user. Then the user can be logged on for hours at a time without blocking anybody else or tying up telco lines, switches and trunks. Further, if it is done right, it won't even block incoming voice calls to the user.

There are numerous suggestions for how this can be done. Usually the telephone company is expected to install some sort of special system that can intercept data traffic before it gets to the telco's central office switch. Northern Telecom's Telesis, a semi-technical journal describing Nortel products, discusses just such a system in Issue 102 (January, 1997). Called the Internet Thruway, and developed jointly with Shiva, it inserts a special line card into the customer's pair which pulls off data traffic and sends it to a data access switch (like an ARPAnet TIP) which then accesses a public or private data (packet) network to the appropriate ISP.

Although an interesting approach, it appears to raise several questions. First, its special line card is between the voice switch and the customer line. Considering that, in spite of all the furor of the moment, only about 10% of residences in the country are using the ISPs and the Internet, and there is no obvious way to tell who is and who isn't, this could, at the very least, introduce administrative problems as well as excessive cost.

Then, tests to maintain lines to the customer, more necessary than ever with today's high speed modems, are presently made via the voice switch (both the Nortel DMS100 and the Lucent 5ESS). Such a test would then have to fight its way through the special line card before it can get to the line itself. ADSL and various exotic schemes to put CATV on the customer's line all face the same problem. But finally, even when Internet Thruway is up and running, it does not appear to do anything to help an incoming voice call reach the a customer busy surfing the 'net.

ISDN to the Rescue?

 To my way of thinking, the correct solution is so obvious that nobody will even mention it: The ISDN Basic Rate Interface. This provides two full duplex circuit-switched B channels at 64 Kb/s each to the customer for voice and high speed data, and a signaling or D channel which can also handle X.25 packet data at 9.6 Kb/s. This D channel could just as easily route packets to Nortel's Internet Thruway data access switch as the new and separate line card; after all, AT&T's 5ESS uses the D channel to connect operator positions to a Directory Assistance data base. Although seemingly slow, 9.6 is fast enough to handle mouse clicks, typing, and no transmissions at all, and a fast download can make a brief circuit-switched connection via one or both B channels. If 128 Kb/s is still too slow, other approaches such as the DirecPC, via satellite, can speed thing up by one or two orders of magnitude, assuming the user has a hard disk fast and empty enough to eat such a meal.

It seems that at least some BRI terminal adapters are not only able to interface voice and data in ISDN formats, but can also use digital signal processing techniques to handle data in "analog" formats as well. Thus, if you want to dial up a direct connection to a friend who has a standard analog modem running at, say, 28.8, the TA reads his signal when he answers and configures itself to map your baseband data as it comes out of your computer directly into the T-carrier patterns needed to come out of the digital transmission system's far end codec looking exactly like modemized data at 28.8. Going the other way, it detects incoming data at 28.8 directly without converting back to tones.

Conclusions

But of course, there's also a problem with ISDN BRI--the service is hard to get and expensive. One gleam of hope turned up when Bell Atlantic, on December 8, 1996, took out a full page ad in the New York Times pushing residential ISDN lines at a discount rate, just in time for Christmas. If the price of an ISDN line can get down to something comparable to two POTS lines, and if the D channel can actually use its X.25 properties to reach the ISPs, and if the ISPs will offer software that can take advantage of modern technology, and if the customers would go for ISDN, we could advance the art enormously. Even after all these years, I haven't given up on ISDN; perhaps it is my hatred of power ringing, DTMF and rotary dials, and preposterously complex modem technology that only six people in the country really understand. ISDN can save us from these things, and allow us to use the digital network and digital switches for something besides analog signals.

What we'll probably get is ISPs accessing the circuit-switched digital voice network with PRIs, using digital signal processing techniques to go from their TIPs directly into the PRI at speeds up to 56 Kb/s (see Michael Finneran's column in the January 97 BCR). This will come out of the T-carrier codec and only have to travel the length of the local loop (which is not subject to the 3.5 Khz limitation of carrier systems) to reach the user's modem.

This vastly more complex digital technology does not take advantage of packet switching and gives us the worst of both analog and digital worlds. We are likely to be stuck with the modem forever, along with the hundred year old power ringing needed to let it answer an incoming call.

[ Top ]  [ Table of Contents ]

Copyright 2006 Lee Goeller. All Rights Reserved.