Local to Wide Area Connections: Access Networks

The link from a LAN, or a collection of linked LANs, to a wide area network is usually accomplished by some variation on telephone company technology. There are two general categories of access networks: carriers and services. We'll talk first about the carriers, which just provide the physical links that create a path from place to place. Other than (in some cases) putting the data into frames, there are no addressing or other software components provided; we could call this U-Switch. Second, we'll review the complete network services that the telcos and their competitors provide. For consistency, think of the carriers as Layer 1 and the services as Layer 2.

Carriers (Physical Links)

There are two basic types of carrier connections—lines provided by the telephone company and those provided by cable television systems. Technically, these are not network services because they don't provide any kind of switching. In the real world, however, that's a little deceptive. The companies that provide telephone lines and cable connections usually also offer connections to the Internet as part of a single package. From the user's point of view, these are clearly services. Still, since the point here is to understand the technology, it's important to separate the physical links provided by carriers from the more complex network services category. Hang in there.

Leased Lines/T Carriers and DS Circuits

A leased line is a physical link that is provided by a telephone company (or Competitive Access Provider; see Tech Talk), but that bypasses all of the telephone system's switches (see Figure 12.4). The user can do pretty much anything with them. The lines can handle analog voice, or digital data, or both. The capacity (speed) of the line depends on the electronics used at the ends (and in the middle, if it is long enough). If, as is almost always the case these days, the circuit is a virtual one, in which the user is leasing not a physical wire but one of many channels on a wire (or broadcast frequency band), then there will be limits on what can be done with the line—you can't, for example, run 45 Mbps over a channel with a capacity of 1.5 Mbps because the electronics that created the channel won't allow it.

An important difference between a leased line (any type) and a dial-up telephone connection (analog or digital) is the way in which you pay for them. Leased lines are charged by distance and by capacity. Time is irrelevant; you pay the same if you use the line 100 percent or 10 percent of the time. This is fair because the phone company has to reserve the capacity all of the time. Distance is important, both in the local and the long distance context. You could pay more to install a leased line that is five miles from the nearest switching office than for one that is two miles away; you will likely also pay a higher monthly charge for the longer distance. A limited exception in pricing for leased lines may be made for government and educational institutions, which are sometimes able to negotiate "postalized" rates (from anyplace to anyplace for the price of one stamp). As for long distance, you can indeed have a leased line that runs across the country, from city to city, or whatever. You have to arrange this with a long distance carrier and you should expect to pay a lot for it. Unlike regular telephone long distance, which has become increasingly insensitive to the distance of a call ("ten cents a minute, anywhere in the U.S."), leased lines usually increase in cost in proportion to distance.

To make it easier to lease lines, the telephone companies lease digital circuits at the same speeds that they themselves use. These are the T-carrier levels. (At one time, it was of course possible to get a leased analog line, but given that the phone network is now digital and given that the services people require are digital, there wouldn't be any point in it.) When the phone companies began to use digital communications, they grouped circuits into multiples of digital voice circuits. We know from Chapter 10 that an uncompressed digital voice channel requires 64 Kbps. Telephone companies found that they could get 24 of these into a twisted-pair circuit (two pairs—one for each direction). This brings you to 1.536 Mbps, to which the phone companies added 8 Kbps for signaling and control. Thus was born the T-1 line with a capacity of 1.544 Mbps. A T-1 line is time division multiplexed, organizing data into fixed-length frames. By the way, you will often refer to T-1 and other telephone system services as being "tariffed." This refers to the fact that charges are officially set not by the carrier but by state public utilities commissions.

A T-1 line can carry not only 24 conversations, but anywhere from 1 to 24 digital data streams, or some combination of both. T-1 multiplexing equipment can combine and separate channels as needed. Signaling is required so that the receiving equipment knows what to do with the incoming data—whether to treat it as one continuous stream, or 24 separate ones, or something in between. Signaling is done out of band—in a separate channel. It's also possible to combine regular dialed voice lines with the data. T-1, like an analog phone line, is just a carrier, a means of formatting data for pushing it down the wire. It doesn't do addressing or error detection/correction. For example, in order to connect two LANs over a T-carrier, users have to add bridges or routers or similar devices at the ends. T-1 lines are usually available to customers who are within 5,000 feet of a telephone company central office or remote switch; longer links are possible, but because they require signal regenerating gear (repeaters), will usually also be quite a bit more expensive. Large customers, those whose needs are expected to grow, may have T-1 provisioned over fiber optics. In this case, very long links are possible.

The next level up from T-1 is T-2, but this is rarely used. Phone companies secure tariffs such that, after you have used the capacity of a couple of T-1s, the next logical step is to go to T-3, which represents 28 T-1s, or about 45 Mbps. The way the pricing works, a T-3, with 28 times the capacity of a T-1, is only about seven to ten times as expensive (this sounds like a bargain, but few people currently use that much bandwidth, and we're talking about going to $7,000 to $10,000 or so per month). T-3 service normally requires a coaxial cable, or a fiber optic line, or perhaps microwave. Before you decide to order one, note that you can get these connections easily in major downtown areas, and in the newer suburban commercial complexes, but not elsewhere. An alternative when you need more than a few T-1s but less than a full T-3 is to get your system provisioned with a full T-3, but only use (and pay for) some fraction of it—"fractional T-3." This is handled by a combination of electronics at either end and a deal with the phone company. There is also fractional T-1, provided in the same fashion, which is more popular for obvious reasons of cost.

By the way, you will see the same circuit referred to as T-3 and DS-3. The difference is that T refers to the carrier (the copper cable and associated electronics) while DS refers to a digital circuit. DS-0 is the official term for a single 64 Kbps channel; a T-1 is 24 DS-0s. DSs and Ts are then synchronized, with T-1 equaling DS-1 and T-3 equaling DS-3. Just to confuse things, there is another level of definition, OC, which stands for optical carrier. We'll discuss this when we cover SONET, but for the moment, note that OC-1, which handles about 51 Mbps, carries a T-3/DS-3 and change.

At the technical level, there are actually three kinds of T-1. The oldest type requires just one twisted-pair but uses a modulation approach, known as Alternate Mark Inversion (AMI), that yields just one bit per Hz. The resulting very high frequency transmission creates a lot of crosstalk in twisted-pair wire, so much so that the telephone companies have to be careful about how close together T-1 circuits are. Standard fifty-pair cables can accommodate only one of these T-1 lines, and they also have to be kept away from other high-bandwidth twisted-pair loops, such as ISDN and the various Digital Subscriber Loops (DSL; see the following section). The maximum distance that AMI T-1 can go without signal repeaters is 3,000 to 6,000 feet, depending on the thickness of the wire (thicker will go farther).

In thinking about these leased lines, remember that while they might traverse many switches and systems, they are inherently point-to-point. The purpose of having a leased line is to get you a connection from A to B. B might be an Internet Service Provider's switch that sends your IP traffic out to the world, or it might just be another computer in your business or organization. Either way, the T-carrier doesn't know and doesn't care; it just frames your data and gets it from A to B. And if it screws up on the trip, someone else is responsible for fixing the problem.

xDSL

The DSL in this technology's name stands for Digital Subscriber Loop, while the x is a variable for different kinds of DSL—ADSL, HDSL, and VDSL. Perhaps by the time you read this there will be xxxDSL. We'll discuss the extant DSLs in turn, but first some general comments. The twisted-pair wire in the telephone company's local loop (see Figure 12.5) can carry quite a bit more bandwidth than ISDN (see the network services section later in this chapter), even quite a bit more than T-1. ISDN's 128 Kbps and T-1's 1.54 Mbps represent the capabilities of the electronics at either end, not that of the transmission medium itself. The actual potential of a pair of wires to handle a signal depends on a number of variables, the most important of which is distance. The electrical signal attenuates (gets weaker) with distance and a weaker signal can't handle the same amount of bandwidth as a stronger one, at least not with an acceptable level of reliability. Remember, a weaker signal means that the changes in the signal (modulation) are much smaller and therefore harder to detect, especially in comparison to the noise that exists in all electronic circuits.

The quality of the wire and its connections are factors as well. Thin wire, or wire with very few twists, can't carry as much data. The way in which wire is bundled also matters; as noted, cables that group tens or even hundreds of pairs are susceptible to crosstalk (a signal leaking from one wire or wires to another). Finally, some techniques that the telephone companies used to improve analog voice circuits (loading) will have the effect of severely degrading the quality of a high bandwidth digital signal. Even when all of these factors are taken into account, there is a lot of good copper above and below us, and some 80 percent of the loops in the U.S. can carry more than T-1 speeds. Interestingly, the percentage of high quality local loops is higher in much of the rest of the world, in part because the cable plant is newer and in part because the U.S. is unusual in the number of very long local loops (the West and Midwest).

Telephone companies have been thinking about DSL technologies for some time now. Strictly speaking, because ISDN is really a form of DSL, we could say they've been working on it for more than a decade. But research on DSL began in earnest in the late 1980s when, as the telcos contemplated further deregulation, they started thinking about the possibility of using their existing cable plant to compete with cable televison systems in delivering movies on demand—also known as video-on-demand and video dialtone. One of the first fruits of this research was the idea that the connection could be asymmetrical—people would download movies but not send them back. So early work concentrated on putting three channels on one wire: 1) a duplex (bidirectional) POTS (plain old telephone service) channel; 2) a fast downstream channel that could handle video equivalent in resolution to that of a VCR; and 3) a small upstream channel that would be used for ordering movies, etc. This video work was in the early stages when trials suggested that the video-on-demand market might not be all that wonderful. Fortunately, the appearance of the Internet provided an alternative reason to get high bandwidth into the home—a use that would also be asymmetrical.

ADSL • The most mature of the DSL standards is known as ADSL for asymmetric DSL. The downstream speed of ADSL depends on the length of the loop; the table below shows the major distance/rate pairings:

ADSL

Maximum distance	Downstream speed
18,000 feet	1.544 Mbps
12,000 feet	6.312 Mbps
9,000 feet	8.448 Mbps

The upstream capacity ranges from 16 Kbps to 640 Kbps. A regular phone channel is available in all configurations. ADSL, like all xDSL approaches, uses a very aggressive modulation scheme (more bits per Hz) that allows for lower frequencies on the wire and therefore, significantly diminished crosstalk. In this regard, it is much superior to T-1.

Implementing ADSL necessitates the usual black boxes at either end of the line, and for a real world use like a connection to an Internet service provider (ISP), requires that there be some path from the end of the DSL line to the ISP. That is accomplished in a variety of ways.

DSL Lite • Remember that one of the advantages of DSL is that no new wire needs to be run to the home or business. This means that the existing physical link has to carry both the voice circuit and the DSL traffic. The traditional way of doing this is to put a black box, a splitter, on the outside of the house in order to separate the regular telephone wiring from the DSL network. The big difficulty with this approach, in industry terms, is that it requires "truck roll"—a visit from a technician. It also requires some investment in the black boxes. But this is a comparatively small problem. When you make a few million black boxes, unit cost falls dramatically, but truck roll doesn't benefit much from economies of scale, and skilled labor costs will always be high. DSL Lite, which is being pushed by an array of computer industry heavyweights and a bunch of telephone companies, is designed to deal with this economic barrier by providing a "splitterless" installation. In DSL Lite, both the phone and the computer plug into the same household telephone wiring. The penalty is lower downstream bandwidth, maxing at about 1 Mbps. On the other hand, tests have shown that DSL Lite doesn't generate as much crosstalk and heat in the telephone company's cable system as even regular ADSL. It looks like DSL Lite will be a common offering, at least for household installations.

HDSL • The next version of xDSL is HDSL, which stands for high data rate DSL. This service, which uses two pairs, offers the same capacity as T-1, 1.544 Mbps in both directions, but goes more than twice as far—12,000 feet. In fact, many phone companies are already providing HDSL when customers order T-1. Because of its more aggressive modulation, it uses lower frequencies and therefore generates less crosstalk. This is important because initial trials of ADSL have already demonstrated that the presence of regular T-1 service in the same cable produces crosstalk that seriously degrades ADSL data rates. Industry leaders are engaged in talks that should lead to a standardized version of HDSL that uses only one pair (SDSL). This should accelerate the trend of using DSL as the mechanism for providing what are traditionally thought of as T-1 services.

VDSL • The final (for the moment) version of xDSL is VDSL, which stands for very high data rate DSL (more naming creativity). This is a very short distance technology.

VDSL

Maximum distance	Downstream speed
4,500 feet	12.96 Mbps
3,000 feet	25.82 Mbps
1,000 feet	51.84 Mbps

The likely use of VDSL is for video-on-demand. The idea here is that the network provider, for example, a telephone company like U.S. West, which was the first to be active in this area, will have a bank of digitized movies in a huge server at its central office. When you select a particular movie, a copy is made and it's played for you—and only for you—over the point-to-point connection provided by VDSL. Since you are the only one watching this copy, you can treat it as you would a VCR—stop, rewind, etc. There are a lot of variations on this approach, but the key is that a high quality digital televison signal requires only some 5 Mbps, so it can fit easily in even the longest VDSL loop. The biggest challenge here is a business, not a technical, one. When Time Warner tried a variation on video-on-demand in Orlando, Florida in the mid-1990s, they discovered that people would rather pay less and go to Blockbuster to rent a tape. Other problems will come with the advent of high-definition video, which will need almost 20 Mbps.

Overall, the DSL technologies face an uncertain future. There is obvious appeal to employing existing cable for new uses such as Internet access and video-on-demand, but the investment will be high. At the moment, it costs about $1,000 to provide ADSL to one subscriber. This includes the black boxes at each end, checking the line, and provisioning (setting up) the service. It doesn't include the network side—the switches, etc. in the telco system that are needed to make it possible to connect with the Internet or other information sources (most DSL implementations use the ATM protocol, which interfaces nicely with Internet connections). Assuming that there are adequate standards in place (not close yet), the usual advances in chip integration could cut the cost of the black boxes by 50 percent or more in a few years. But this presents the ususal chicken and egg problem—can you get enough users at the higher price to justify the scale to bring manufacturing costs down? And, most important, will people pay enough on a monthly basis to amortize the investment? The answer so far appears to be yes.

Cable Modems

Now we'll talk about some folks who do have great technology but—at least until AT&T's purchase of industry leader TCI—didn't have the cash to take advantage of it. Television cable systems employ coax, which has bandwidth potential far in excess of the 450 to 750 MHz normally used, and they are eager to take advantage of their existing plant for Internet services (not to mention telephone services and video-on-demand). A typical television cable plant includes 75 to 125 6 MHz analog channels. If you take one of these channels out of service (e.g., The Curling Channel), you can put a pair of modems on either end, stir in about five or so bits per Hz of modulation and you have an awesome 30 Mbps to deliver to nerds in the 90 percent or so of residences and businesses that are "passed" by cable (have access to it, don't necessarily subscribe) (see Figure 12.7).

Unfortunately, it's not that easy. One problem is that many cable networks were pretty sloppily put together and have lots of poor quality connections (in and out of the home) that will degrade the signal. The external aspect of this will be rectified as cable companies complete their periodic rebuilding, this time with more emphasis on quality. The in-home side is likely to cause a lot of headaches; few user-installed cable taps are any good.

Another big problem is that only 20 percent or so of cable networks have any upstream capability, and most of that is not of high quality. Cable systems that do provide uplink capacity have mostly done so with bandwidth at the low end of the frequency spectrum (the lower channels). This is where there is a lot of ambient radio frequency interference, including CB radio, amateur radio, and shortwave. Prior to the popularity of the Internet, the few cable companies that had a return path used it for nothing more than low data rate applications, such as pay-per-view or, in some bold experiments, letting couch potatoes push buttons to vote on their dream date. For these heady intellectual pursuits, a noisy, low data rate channel was OK. It's not so great for Internet access, though.

Internet-over-cable systems propose three ways of dealing with the upstream problem. One is to put the uplink elsewhere—the user employs a normal modem/analog phone line for the upstream side. The second is to bite the bullet with the low frequency channel and go ahead with a modest modulation rate that resists noise even as it delivers much lower data rates. The no-return-channel option appeals to companies who don't have upstream capabilities and don't want to rebuild to add it. Others are using slow speed uplinks and some are investigating approaches to modulation that better resist noise (CDMA, which stands for Code Divison Multiple Access, is a technology under active investigation; it works well in cell phones).

The third and most desirable approach, of course, is to have the cable companies totally rebuild their networks with fiber as close as possible to the customer and with new electronics that allow additional, high quality channels for data. Better still, junk the analog stuff and make the whole thing digital. Unfortunately, these advanced solutions not only require rebuilding more of the network, they also assume replacing all the set-top boxes as well. This poses a big financial problem for the financially challenged cable companies. Costs would be lower if the cable companies could develop a standard digital set-top architecture. CableLabs, which is a research and standards organization funded by cable providers, has been hard at work on such a standard. The biggest challenge to this process comes from Hollywood. The moguls are very concerned about the advent of a new generation of cable boxes that allows direct connection to digital televisions and recorders. In such a scenario, technically the most desirable one, it will be easy for people to make perfect copies of movies. As this is written, there is hard work but little agreement on a scheme to prevent this.

Hot Wired Connections There's another possible solution to the "last mile" problem—the electrical grid. Power companies have found that it's possible to carry data over the lines that bring electricity to homes and businesses. Since power lines are, to put it mildly, an electrically noisy environment, a lot of error detection and correction is needed, and this limits bandwidth. But the real issue, not resolved yet, is whether the available capacity can be cost-effective in the surging market for home bandwidth.

A cable modem network uses the tree and branch topology because that's how cable television networks are set up. At the user end, there is a cable modem that connects on one side to the PC (usually by Ethernet) and on the other to the cable. The cable then connects to other cables and ultimately to a switch of some sort. The key point here is that the part of the network that extends from the customer premises to the first switch functions like a bus-based LAN—it has shared bandwidth (compare to the star topology of xDSL and other telephone-based approaches). Depending on the situation, any given user has a lot of potential competitors for network access—120 to 500 in current systems, with the average being about 200. Where I live, in Ohio, this isn't much of a problem; my neighbors mow their lawns and watch TV. On the other hand, at my sister and brother-in-law's place in southern California, where every house has at least one fanatical cybersurfer, competition for bandwidth is fierce. The solution to this problem is for the cable companies to push fiber optic multiplexers and switches closer and closer to the customer premises, making for an ever smaller number of users sharing the same bandwidth. And, by the way, users need to protect their computers from intrusion—the shared nature of cable networks means that nosy neighbors could get access to your drives and printers.

Network Services

The communications systems described in this section—dial-up analog lines, ISDN, Switched 56, X.25, Frame Relay, and Switched Multimegabit Data Service—are different from the dedicated line/carrier systems described earlier (leased digital lines, xDSL, and television cable lines, summarized in Figure 12.4), which are just means for putting data over a wire. Some network services are dumber than others; dial-up phone lines, ISDN, and Switched 56 only provide the means for making a point-to-point connection. They are sort of intelligent Layer 1 systems. By contrast, the other network services that we will discuss, X.25, Frame Relay, and Switched Multimegabit Data Service, include some error detection/correction and flow control. These operate (more or less) at Layer 2.

Dial-up Analog Lines

The standard telephone line from a home or business is analog and limited in bandwidth to about 4 KHz. To use these lines, computers have to have modems at either end. The modems convert the digital data coming from a computer to analog, modulate (code) it, then send it over the line. The modem at the other end decodes the data and converts it back to digital. Modems, which work through the dial-up telephone system, create switched circuits (see Figure 12.8). The modems will provide flow control and error detection/correction, but don't handle any addressing and don't know or care what kind of data they are carrying.

The modulation (coding) employed by modem vendors has become ever more sophisticated, employing amplitude, phase, and frequency modulation all at once. The result is that as much as 33.6 Kbps can be stuffed into 4 KHz. Since the bandwidth actually used by the telephone electronics is about 3.3 KHz, modems are now getting 10 bits per Hz. This is very close to the theoretical maximum. In the real world, of course, noise in the line makes this aggressive modulation infeasible, and the modems often drop back to a slower speed.

ISDN

ISDN (Integrated Services Digital Network) is the digital equivalent of the analog phone line. The standard form of ISDN, called Basic Rate, includes two 64 Kbps channels plus another 16 Kbps for signaling, on a single pair of wires. This means that ISDN has a capacity of 144 Kbps (full duplex). Since normal analog lines only carry the equivalent of 64 Kbps, ISDN produces what the industry calls pair gain. This phrase describes a technology that allows a pair of wires to gain capacity.

ISDN normally functions as a switched service. Unlike a leased line, it isn't open all the time. Instead, just as with a regular voice phone call, you make a call to the telco switch to set up a circuit and it lasts until someone does the equivalent of hanging up. Needless to say, there is no point in making a data call via ISDN if the party called doesn't have ISDN service as well. ISDN is provisioned on only one twisted-pair and can be as far as 18,000 feet from the telephone switch. Since most homes and businesses (about 90 percent) are that close and have at least one pair available, existing wiring can be used in the vast majority of cases.

ISDN is flexible. The standard ISDN service, Basic Rate (BRI, for basic rate interface), is known technically as 2B+D, because it has two bearer channels of 64 Kbps and one signaling channel (the D) of 16 Kbps. The D channel is actually a packet-switched link that is always open to the switch. When a phone call comes in from the switch, it is signaled over the D channel; when the phone is picked up, the B channels are opened for the conversation. The same process can be used for data, with a variety of possibilities, the most important of which is that the full 128 Kbps (2 × 64) can be used at one time (in one direction). ISDN is full duplex—these capacities are available in both directions at the same time.

Installing ISDN • At the telephone company end, installing ISDN is fairly straightforward. First, the central office switch has to be ISDN-capable, which means that its software must be able to talk to the ISDN system. Second, the interfaces to the incoming line, line cards, have to be swapped from analog to digital. Connections can then go through the switching system just like any other 64 Kbps voice circuit. At the home or office end (customer premises in telco jargon), the situation is more difficult. You need black boxes. One kind of box is used to connect a computer to the line (the wall jack); the link from this box to the computer is normally Ethernet. Both the computer and the ISDN line are digital, but manipulation is required both to fit the computer's data signal into the ISDN format and to manage bandwidth. By definition, an ISDN line can handle voice, and if you happen to have a couple of ISDN-compatible digital phones sitting in your attic, that's all you need. More likely, though, you have a bunch of "free gift" analog phones that you don't want to give up. No problem, you can buy another black box that will allow you to connect these. An arcane feature of ISDN is that it can allow one physical line to carry a bunch of different phone numbers. This will allow people to have separate numbers for voice, fax, and computer. There is usually an extra charge for this feature, though. Phone numbers are in short supply.

Options and Costs • There is another kind of ISDN, Primary Rate ISDN (PRI), which has the same bandwidth as T-1, 1.544 Mbps, and also requires either one or two wire pairs, depending on the local service situation. Since both Primary Rate ISDN and T-1 can carry switched-voice conversations as well as data at varying rates, the principal difference between the two is that ISDN connects directly to the telephone company's switched network and T-1 is just a vehicle for getting bits from A to B—it has no switching capacity. PRI is more flexible, but usually more expensive.

Depending on the carrier, you will not pay a higher monthly charge for a longer ISDN line. Local calls will be billed at either a flat rate or based on time of use, again depending on the phone company. ISDN long distance calls are like telephone long distance.

ISDN is a simple enough idea. When first developed, about 1984, it also seemed feasible. Telephone companies were well along in the process of converting their long distance systems to digital circuits and the move toward all digital local switches was rapidly gaining momentum. Converting the "local loop" to digital seemed a logical next stage. That was the theory, anyway. The fact is that ISDN didn't happen on the scale intended, and it appears it never will. Interestingly, there were no technical barriers; the problems had to do with standards and market.

Problems with ISDN • Telephone companies, which have to connect with each other to provide national and international long distance service, have a good record of cooperation on technical standards. As an example, while the European equivalent of the T-carrier is different from the American (it has 32 channels and carries about 2 Mbps), the signaling standards are the same and the two can interoperate easily. ISDN, on the other hand, is a local technology and interfaces directly only with the company's own equipment (the central office switch). Thus, while international standards were developed for ISDN, as often happens they didn't specify everything in full detail—it usually takes quite a bit of time and experience to do that. So different companies implemented ISDN differently. This fragmented the equipment market and helped keep prices high. These incompatibilities are now largely solved, but they impeded progress for many years.

The market problem is more serious. When ISDN was first deployed, the Internet didn't exist except for a handful of users with esoteric interests. Data users at that time were almost exclusively large companies who met their needs with leased lines. Small businesses used dial-up analog lines. Most businesses and nearly all individuals had no data connections. ISDN's switched 128 Kbps was too little for the leased line users and too much for everyone else. Also, it was very expensive and not widely available (remember that availability is two-edged—it doesn't matter if you have ISDN if the system you want to connect to doesn't). For a decade or so, ISDN was like the weather; it was much discussed but little was done about it. With the exploding popularity of the Internet, and especially of the Web, fast modems quickly seemed slow and phone companies moved to expand ISDN offerings. Pretty much universally, they did a bad job of this.

Because of their long experience as monopolies, phone companies have a record of not understanding marketing (they do know how to advertise, but that's just a part of the deal). Indeed, providing a complex product in a way that "delights the customer" was not in their usual bag of tricks. And from the telco point of view, ISDN was not just any service. Making it happen required intra-company communications on the part of groups that didn't know each other existed. Horror stories about trying to order ISDN and getting a confused answer, followed by a promise of installation that took months to fill, followed by a nonfunctioning system, were commonplace. ISDN pricing has also been a problem. A number of telcos initially promoted the service by offering a monthly flat rate for local calls. When they discovered that many, even most, of their users were Internet junkies who left the line open 24 hours a day, they expressed shock and tried to change the pricing. Needless to say, this didn't lead to a high level of customer good will.

After a few years, the phone companies got their stuff together and made ISDN reasonably quick to order and install, and somewhat affordable. Unfortunately, at this point, ISDN's 128 Kbps was nice, but not exciting. It looks like cable modems, xDSL technologies, direct broadcast satellite, and wireless local loops, all of which offer far more capacity, will keep ISDN from being a factor in the access market.

The gnomes at the International Telecommunications Union (ITU, née CCITT) are working on a high-speed extension of ISDN called Broadband ISDN. This will be a protocol structure that sits on top of SONET and ATM, and goes to about 600 Mbps, or something like that. Don't hold your breath. History suggests that once complex schemes like this make it through the standards process, something better will have gained a strong lead in the market.

Switched 56

Switched 56 is a bit of an in-between strategy. A pure data service, it's cheaper to install than a fractional T-1 and has the advantage of being dial-up, so subscribers pay only for the bandwidth they use. On the other hand, it has less capacity and less flexibility than ISDN. Given that it can't command much market scale, it probably won't be around long.

X.25

X.25 is little known by the general public, but it is a very widely used "back end" technology. X.25 is responsible, among other things, for automatic teller networks used by banks, for credit card verification services used by hundreds of thousands of merchants, and for some low-speed, low-traffic terminal-to-host connections. There are a number of network providers. The local end, including lines and terminating equipment, is normally maintained by telephone companies. Once at the telco's central office, packets flow into a broader switching network that could belong to the telco itself, to a long-distance carrier, or to a third party like CompuServe. Charges to the customer depend on a variety of factors and may come from multiple sources (the local connection provider and the network provider). Permanent circuits are likely to be charged according to packets sent, while temporary circuits will be charged on a time-connected basis or perhaps some combination of the two. From the customer's point of view, the advantage of X.25 (or other carrier-switched services like Frame Relay) vs. a network built from leased lines and locally maintained switches is that the user doesn't have to be in the networking business. The service provider handles all the plumbing and is responsible for being sure that packets get to where they are supposed to go. In addition, since other businesses are also connected to the same network, it is possible to have an array of links beyond your own organization.

X.25 is a packet-switched network, one that uses a call setup procedure to establish virtual circuits, either permanent or temporary. Once the path through the network is established, individual packets carry a virtual circuit identification rather than full destination and source addresses. Because each packet (usually) follows the same path, packet sequencing isn't a problem. On the other hand, X.25 is especially serious about error detection/correction and flow control. Each receiving switch checks each packet for errors and returns an acknowledgment to the sender before another packet can be sent. Individual switches also handle flow control. One reason that X.25 does error detection/correction and flow control in the switches rather than at the ends of the network is that it was created for a world in which the devices at the ends—terminals—were too dumb for these tasks. Another reason for X.25's paranoid approach is its heritage as a technology for old-fashioned copper wire telephone connections, in which errors were quite common. X.25 normally runs on a line set up for 56 or 64 Kbps, though the extensive error checking means that throughput is quite a bit lower than this. While still widely used today for low-speed connections, X.25 has been bypassed for advanced data exchanges by its younger, sleeker sibling, Frame Relay.

Frame Relay

Frame Relay is a very hot technology that is offered by telephone companies (mostly) as the logical step up from X.25. It is in many ways an enhanced version of X.25. As with X.25, the telco provides the user with a dedicated line and terminating equipment that connects to an internal computer network (usually to a router) (see Figure 12.9). Where X.25 uses its own physical layer, Frame Relay sits on top of an existing carrier—Switched 56, ISDN, or T-carrier. Frame Relay frames (which can be as long as 4,096 bytes), like X.25 packets, are switched through virtual circuits following a call setup process that replaces packet addresses with virtual path identifiers. With minor exceptions, these circuits are permanent virtual circuits that the user establishes by agreement with the telco. In other words, you designate the connections that you will want to make (e.g., to remote offices and to an Internet Service Provider) and these connections are put into routing tables in the switches. Switched virtual circuits (equivalent to dial-up connections) for Frame Relay are a relatively recent innovation.

As with ATM, to which it is closely related, with Frame Relay, the user's permanent virtual circuit connections are programmed into the appropriate switches so that when a frame leaves a network site with the appropriate circuit identification in its header, the switches in its path recognize it and know exactly where to send it. Not only do the switches know where the frame should go, they also know how much bandwidth it can have. The user and the telco have agreed in advance on the average circuit speed, which is called the committed information rate, and a burst rate, which is the amount over the committed rate that the circuit can go for brief periods. Some Frame Relay providers also offer bandwidth-on-demand, which means that if the network has excess capacity, users can temporarily get a data rate higher than theburst rate by requesting it from the carrier. Of course, the pipes into and out of the network (the leased lines) dictate the maximum end-to-end speed—you can't get T-3 out of a T-1.

Another way in which Frame Relay is different from X.25 is in error correction and detection—X.25 does a lot and Frame Relay does almost none. Frame Relay detects, but does not correct; "bad" packets are simply thrown away. Frame Relay, together with other protocols that don't do error correction or flow control, is called a fast packet technology. When it was first marketed in the late 1980s, most of the telephone network consisted of high-quality digital circuits, including virtually error-free fiber trunks. In this environment it doesn't make sense for each switch to verify each frame; instead, the upper layer protocols employed by the user (e.g., TCP) can deal with the relatively rare problem of corrupted or missing packets by asking for retransmission. While Frame Relay switches do less work than their X.25 counterparts, they are more intelligent. The bandwidth-on-demand option is one example; multicasting is another. X.25 allows only a one-to-one transmission, but Frame Relay allows one-to-many connections.

Frame Relay specifies how a network is accessed, not necessarily how it works internally. Thus, telcos increasingly are using Frame Relay as an interface to internal networks that use other protocols, notably ATM over SONET, to move information around inside their "cloud." Once user data, chopped into cells and stuffed into slots, reaches the other side of the cloud, frames are recreated and sent to the addressee (which will then strip off the Frame Relay frames, look inside to find the TCP/IP address of the ultimate destination, then put the packets inside Ethernet frames). A lot of electronic slicing and dicing goes on in networks, and it's really amazing that it all works.

Frame Relay has a variety of uses. Many companies employ it as an alternative to leased lines for connecting remote LANs. The T-1 speed is the likely choice for this use. Frame Relay's big advantage over leased lines is that one site needs only one line to connect to multiple other sites—e.g., both to remote offices and to the Internet. The alternative, multiple leased lines with a user-provided switching center, is likely to be much more expensive. Because Frame Relay's variable length frames can introduce latency, and because its switches aren't designed to manage quality of service constraints, it is not considered suitable for video or voice. However, in the tradition of people adapting the tool they have to whatever task is in front of them, some vendors are now trying to add quality of service elements as well as voice to Frame Relay.

Frame Relay tariffs are normally a composite of the cost of the physical connections to the carrier's network and a per-frame charge. A low speed/low use link might run a few hundred dollars a month. A highly utilized T-1 connection could be several thousand. Remember, though, that in a fast-changing, competitive environment, both the variables and the rates are subject to frequent change.

Switched Multimegabit Data Service (SMMDS)

SMMDS, like X.25 and Frame Relay, is a data networking service offered by telcos to businesses. It is a much higher speed system, provisioned at rates up to 45 Mbps, though 1 Mbps or so is most common. At the physical level, SMMDS uses something known as a Metropolitan Area Network (MAN), which you can think of as a city-wide LAN run by a telephone company or Competitive Access Provider (CAP). We won't go into the technology here, just note that MANs use a time slot organization, have a dual bus rather than a dual ring topology, and are also designed to have a high level of resistance to failure. For the upper layers, SMMDS uses a cell-based switching approach that is very similar to ATM; its packets are, in fact, the same as ATM's. The real question with SMMDS is whether it is a viable alternative to ATM itself. More and more telcos and competing access providers are offering ATM services directly. Given the market scale of ATM, especially when combined with SONET, it seems that SMMDS has little long-term potential.