If you're clear about your needs and priorities, choosing the best network services for your company doesn't have to be mind-boggling.
| Illustration: Daniel Guidera|
As the 20th century ends, most organizations find themselves involved, one way or another, in upgrading or extensively revising their computer networks. Chances are, this trend will continue for at least the first decade of the coming century.
This will be a time of tremendous opportunity for creating networked applications that can save or make organizations millions of dollars. It will also be a time of turmoil, in which firms can easily mismanage their networks and waste much more of their network budgets than they could have expected.
We all know the short-term reasons for the current stress on networks. Among them is the rapid rise of the Internet and its opportunities for low-cost global communications, electronic commerce, and potential mergers between voice and data communications. But today's networking trends are merely part of a long march toward more available, more flexible electronic communications.
If we look at pictures of city streets in the first decades of the 20th Century, we see that they are jammed with wires--thousands of wires on hundreds of telephone poles. Our need to make our computers communicate is a modern parallel to our grandparents' need for better telephone communications.
Networking trends are unstoppable, and somewhat frightening. At the turn of this century, a company could get by for a decade or so without a telephone. Now it's likely that in developed countries, companies will go out of business in a year or two unless they intelligently use the Internet.
But that doesn't mean all new networking approaches have to be adopted uncritically. For the foreseeable future, IT's job must be to make sure networking applications and services contribute to overall organization performance. This means selecting the right network services to carry the flow of information. And, despite rosy predictions of virtually free WAN communications within a couple of years, it means finding ways to keep expenses down.
The fourfold path
Making the right network services selection isn't an impossible task. In fact, the issues involved are relatively simple.
Match your network configurations to your traffic.
Any organization reviewing its network services must, as a first step, match its network configurations to its overall flow of information.
Note the word "overall." It may well turn out to be economical to waste money on a circuit or two, or even give some parts of the organization poor communications services, as long as the bulk of communications is handled economically and efficiently.
Plan your networks for the future.
The second step is to plan for the future. The trends are already clear. For starters, organizations are using more bandwidth than ever before; per-organization use of WAN bandwidth is increasing at 25% to 40% per year, according to various surveys.
And don't bank on WAN communications being virtually free anytime soon. Technical savants say that by 2001, optical communications technologies such as Dense Wave Division Multiplexing (DWDM) will greatly drive down the cost of wide area networks. This is indeed possible, but the introduction of lightwave communications didn't drive down prices overwhelmingly--so why should DWDM?
Other trends to bear in mind:
Organizations are increasingly relying on the Internet.
Organizations are finding that they need secure and robust communications.
And organizations are preparing for increased access by mobile and home workers.
Take stock of what you've already got.
The next logical step is to check the economics and technical suitability of the network services your company currently uses. It's a truism that communications facilities, once installed, are rarely removed. So your firm likely has legacy technologies along with the latest new technologies.
Older technologies sometimes can't be decommissioned without changing the applications involved. Sometimes it's not cost-effective to make the switch. In addition, many of those older services and technologies work pretty well. Don't abandon them without careful thought.
For instance, there are signs of increasing competition on long-haul and, in the future, local-access communications. New carriers with a flood of new capacity are already driving down prices. But don't forget the price and service advantages of signing long-term contracts with the established carriers.
Migrate to new technologies and services where it makes sense to do so.
| It's all about priorities |
Here are the main questions to ask when making a choice among leased lines, frame relay, and ATM:
Are you choosing a network service for an individual connection or for the entire network? If just one, it's probably best to stick with whatever service you're using for the rest of the network.
Can you create a hybrid solution (leased lines for some connections, ATM for others, for example)? If you can create a hybrid, consider whether you will want to keep it, or move to a single service at a later date.
Is the overall network mainly many points to one point, many points to many points, or a few points that can be connected in a triangle, square, or rough circle? Your choice will be determined more by the configuration than by the cost, because leased lines are often cheaper than packet services for short distances and point-to-point connections.
Do you plan to use individual links nearly constantly, and do you require a guaranteed amount of bandwidth? If so, leased lines are the clear choice.
Does your network require guaranteed security? Leased lines again have the advantage. Security specialists recommend encryption on any shared service.
How much bandwidth do you expect to use on each circuit or virtual circuit for the foreseeable future? It's advisable to buy more than you need because network traffic is growing so rapidly.
Do you require backup services? Leased lines can only be backed up by other leased lines or dial-up services. frame relay and ATM lines can be backed up interchangeably through multiple service carriers . (for instance, you could back up an AT&T frame relay service with one from Sprint.)
What is the total cost, including local access and installation for a specified time (usually three years) for each alternative? --Paul Strauss
Migrating to better approaches means weighing the pros and cons of various network services, while bearing in mind that any new service or technology must support the organization's business objectives.
The choice of a service shouldn't be based on whether it's theoretically cheaper or better than another, but on whether it truly and dependably is. In the case of frame relay services, carriers allow users to "burst" over a Committed Information Rate, and most corporate networks rely on using the burst rate constantly. But the situation could be temporary. If carriers restrict over-use of the "burst" capability, the attractiveness of frame relay as a voice/data service will fade.
Users are constantly faced with a choice between leased lines and frame relay, and sometimes they must choose between frame relay and Asynchronous Transfer Mode (ATM) (see sidebar, "It's all about priorities").
Here's how those services stack up.
Leased lines are the clear choice for organizations that plan to use individual links nearly constantly, that have networks covering short distances, or that require a guaranteed amount of bandwidth. If you have a lot of traffic between Atlanta and Chicago, for example, it would be far better to lease a 56Kbps or T1 (1.544Mbps) circuit instead of making constant modem calls or Integrated Services Digital Network (ISDN) calls. Leased lines also are a good choice for organizations whose networks require guaranteed security.
It's relatively easy to compute the cost/benefit tradeoffs among different kinds and different speeds of leased lines. But in many situations it's not clear whether firms should take leased lines at all.
Of the two kinds of leased lines, local-access circuits and long-haul circuits, there have been some price declines in the former over the last 18 months. These declines have been caused by new competitors offering Internet access services, often by reselling the service offerings of the entrenched local service companies.
In general, local-access price declines have ceased and prices are tending to rise slightly in major markets, says Thomas Jenkins, a senior consultant with TeleChoice, a consulting firm in Owasso, Okla. Although prices of local access vary significantly by metropolitan area, the list price of long-haul leased lines varies little by area or by carrier and has not changed substantially in years. If anything, long-distance circuit prices are tending to increase slightly, Jenkins says.
Most companies find that leased local-access and long-haul circuits are terribly expensive in the absolute sense, consuming a major portion of the annual WAN budget. A T1 between Atlanta and Boston runs around $5,000 a month these days. And naturally, the expenditure seems overwhelmingly onerous if circuits are not constantly being used.
New technologies to some extent ameliorate the wastefulness of taking leased lines and not using them for periods of time. More significantly, they deal with the problem of multipoint connectivity. Today's digital leased lines are only point-to-point circuits (unlike analog circuits, which offered multidrop options). This means that a circuit runs only from Atlanta to Chicago, for example. It's possible to connect Atlanta to Chicago to Boston by using multiplexing technology, but that takes two leased lines. Four points take three leased lines, five four, etc. More expensive still is the creation of a matrix network, in which all points connect to all other points. Connecting five cities in a full matrix requires a star configuration of 10 leased lines.
An additional drawback of leased lines is that they can only be backed up by other leased lines or dial-up services.
Packet services, which make more efficient use of circuits, evolved as an answer to the problems of leased lines. Packet services often are cheaper than leased lines because the carrier can aggregate the packets from various users on the same circuits.
Frame relay was created by the telecommunications standards bodies because by the mid-1980s, X.25 was considered a slow network transport protocol. Frame relay allows for faster transmissions than X.25 because it doesn't require error checking at each switch. Recent survey data from Datamation and SG Cowen show marked increases in interest and use of frame relay.
Many organizations think they save money by putting voice on frame relay services. Doing this can save money, but only when there is unused capacity on a subscribed frame relay permanent virtual circuit (PVC)--which is essentially a connection. With the exception of international calls, which are often tremendously expensive, it rarely saves money to put voice on a data service if you have to pay to increase capacity of that service.
Frame relay can provide up to T1 speed in the United States and E1 (2Mbps) in Europe, and it has been tested at 45Mbps. But few carriers offer such fast frame relay services because those speeds are considered to be ideal for ATM, which apparently has no top speed.
ATM can and does carry voice and data. Much of the dialed voice traffic in the United States already travels over carrier ATM switches, and recent Datamation/SG Cowen surveys indicate ATM use is soaring. But ATM does not yet carry much corporate voice/data traffic.
This is surprising because communications carriers are clearly using ATM to carry both voice and data. If carriers use ATM for voice and data, why don't corporate users? Largely because corporate users cannot obtain the economies of scale the carriers have.
Also, carriers use ATM differently from the way corporate users want it. Carriers use an ATM format known as the constant-bit-rate format for voice, and another format, the variable bit rate, to carry user data services. Corporate users were hoping to obtain a single ATM service for both voice and data. Having to buy two different formats raises the cost and diminishes the flexibility of ATM.
Sorting it out
It's usually easy to compare the cost of individual links--a T1 leased line from Atlanta to Chicago versus a frame relay service with a T1-speed Committed Information Rate between the same two cities, for example. And it's easy to compare a simple multipoint leased-line network with the same connections made through frame relay. Frame relay configurations generally cost between 20% and 35% less than comparable leased-line networks.
ATM wide area networks, which usually start at speeds of 45Mbps at the slowest, often cut the monthly cost of leased-line configurations in half. More importantly, they allow multipoint configurations that were never possible with leased lines.
It's also relatively easy to decide between frame relay and ATM. At speeds of less than T1, you must take frame relay (ATM is not offered). At speeds over T1, you usually must take ATM (frame relay is rarely offered). At T1 speed, you can choose between the two.
(There is often a misleading discussion of the so-called "cell tax," meaning the amount of overhead on ATM networks. This amount can run as high as 20% due to the high proportion of cell headers and inter-cell spacing bits to payload. But the 20% overhead only affects a network choice when there is a direct comparison. If a frame relay and ATM PVC both have a T1 Committed Information Rate and both cost $1,000 a month, then the frame relay network will deliver higher throughput because of its lower overhead. In general, however, ATM's high overhead doesn't cost users more than an asynchronous packet service because the fact that it's synchronous helps to reduce the cost of switches. Many frame relay switches are actually ATM switches.)
While it's sometimes easy to choose between frame relay and ATM, in general it's by no means easy to price out comparable networks, largely because of the deregulation of telecommunications. ATM and frame relay are no longer sold as tariffed services, which means that their prices are not listed. Indeed, it's hard to say that there are fixed prices for these services from many carriers. Most carriers set a price by negotiation, which means that large networks are cheaper than small ones, and long-term contracts are cheaper, per month, than short-term contracts.
Moreover, there are signs that some carriers will offer more attractive prices on a data service if the deal also includes the customer's voice services. If deals are made on this basis, it's meaningless to attempt to compute the price difference of ATM over leased lines.
Recent figures show rapid declines in the use of leased long-haul circuits and rapid increases in the use of frame relay and ATM. All three services are usually offered by major long-distance carriers in all metropolitan areas. With the rise in Internet popularity, T1 and fractional T1 circuits are often being used as access circuits, and these facilities also are available in virtually all metropolitan areas. Lagging are local frame relay and ATM access services, which in some cases are marginally less expensive than leased access services.
Other network services
While much of the attention in wide area networking has focused on such services as frame relay and ATM, there are alternatives that can be highly cost-effective for small branch offices. Two technologies to consider are very small aperture terminal (VSAT) services, communications satellite broadcasts received by relatively small dish antennas (usually about four feet across), and ISDN, a dialed digital service that was intended to replace today's analog telephone system. VSATs can be highly cost-effective when there are scores or more branch offices that tend to receive most of their information from headquarters. They are not terribly effective in two-way mode.
The basic-rate ISDN service was originally designed to give all users two 64Kbps digital circuit-switched channels and a 16Kbps packet-switched channel, and to be used either as a telephone service or for data transmission. Unfortunately, because of variations between services, ISDN rarely offers all of these capabilities and virtually never can be used to replace a telephone line. ISDN installations are notoriously difficult, and there are enormous variations in pricing for these services, from a fixed rate of $17.90 a month in Arkansas to $249 a month plus per-minute charges in most Bell Atlantic areas. Still, this is a lot less than the cost of leased lines, which may cost $300 a month for 56Kbps of bandwidth or $500 for a 128Kbps circuit.
While ISDN is by definition a switched service, some ISDN circuits are being sold as point-to-point connections, or the equivalent of a leased line. These can be attractive for access to the Internet, but they're not available in all areas.
New high-speed access services
With ISDN a disappointing option, most organizations are forced to access the Internet or their private data networks using leased lines. In some cases, access is possible using frame relay services.
Since T1 leased lines cost around $1,000 a month plus $1,000 for installation, there are hopes for new services that will be considerably cheaper. The two technologies that are most discussed for doing this are Digital Subscriber (fixed) Lines, which is usually called the xDSL technology these days because of the many speed options, and data over cable TV services, called cable-modem services. Unfortunately, both of these services have been rolled out very slowly, and it's unlikely that either of them will be a major competitor to leased lines for many years.
However, for workers in home offices, both services could be relatively inexpensive. Bell Atlantic has just announced its planned xDSL services, which will be "asymmetrical," meaning that they will have relatively low speeds of sending data but will allow downloads at up to 7.1Mbps. For this service, Bell Atlantic is charging $189.99 a month (which is a lot of money for the average home user, but cheap in comparison to the cost of a leased line). Cable modems offering a download speed of 1.5Mbps are offered for around $40 a month.
On the high end, the latest trend is for carriers to lease OC-3 (155Mbps) and OC-12 (622Mbps) fiber-optic access and long-haul circuits. Prices are rarely listed, but they are likely to be in the tens of thousands of dollars a month. Like T1 circuits, these are simply very high bandwidth point-to-point leased lines. However, they can be used for both voice and data, and for many large companies, they cost less than taking multiple 45Mbps circuits.
Virtual private networks
Because corporations already have Internet access for their Web servers and increasingly use the Internet for electronic mail, there is growing interest in using the Internet for data transport as well. This has many theoretical advantages, the main one being that the Internet, unlike ATM and frame relay networks, is essentially one big network made up of thousands of carriers' backbones. This being the case, you don't have to worry about whether the carrier offers service in Pomona, Punta del Este, or Penang--any global point can be reached.
There are plenty of misleading calculations showing that secure Internet data networks, known as virtual private networks (VPNs), are cheaper than ATM or frame relay networks. (VPNs are considered secure because they use a form of encryption known as "tunneling," but this only prevents outsiders from reading the communications; it doesn't tunnel packets through the network in the sense of giving them priority or additional robustness.) The rationale for asserting that VPNs are cheaper than frame relay or ATM is that with frame relay and ATM services, you pay for access and then you pay for the long-distance service, usually several hundreds of dollars a month per permanent virtual circuit (PVC). With a VPN, you pay only for access; the Internet backbone is said to be free.
This is not true, of course. Nothing is really free. If you pay for Internet access, you pay for the wide area backbone services used. If they are congested and must be upgraded, the carrier will pass the charge on to you. In fact, Internet backbones are badly congested and often suffer from extensive packet loss.
Also, VPNs are slower, less robust, and more vulnerable than other data services (see text box, "The downside of virtual private networks"). So, although Internet VPNs are cheaper than other data services, in terms of their ability to deliver data, they are often less cost-effective.
There are many interesting technologies proposed to improve the Internet's congestion and increase bandwidth. One of the most promising is Multiprotocol Label Switching (MPLS), which allows carriers to toggle between their current networks--based only on the Internet Protocol--and ATM backbones, and it allows large IP backbones to have better flow control. However, MPLS is not likely to appear for a year at least.
In short, VPNs appear attractive, but in most cases they're not as cost-effective as frame relay or other network services. There are a number of important exceptions, however. An Internet VPN is an attractive option for a corporation that must put together a truly global network but doesn't require high-speed communications to all points. Leased lines and packet services often are highly expensive in developing countries.
A second attractive use for a VPN is to replace modem banks and other dial-in equipment, such as remote-access servers. In this case, users dial in to the Internet and access the corporate VPN by giving passwords. This approach is usually much cheaper than having to maintain multiple telephone lines for dial-in modems, and it eliminates the often heavy expense of maintaining staff who are specialists in modem access. //
Paul Strauss is an independent networking writer and consultant who formerly worked as a senior editor with Datamation and held various editorial positions at Network World and Data Communications.
A brief history of networking services
First, there was Samuel Morse. Actually, first there were smoke signals. Then there was Samuel Morse.
Electronic networking services date to Morse's telegraph in the 1840s. The cost was horrendous. The first messages sent by trans-Atlantic cable in 1866 cost $150 in greenbacks (or $100 in gold) for 20 words!
Not many organizations use telegraph any more, but some financial-services companies still use its successor, telex, for certain legal purposes, according to Unitel (http://www.unitel.nl), supplier of telex hardware and software. Users number 1 million and are concentrated in the banking and shipping industries, Unitel says.
(The telegraph's great-great grandchild, e-mail, emerged as a popular technology in the 1980s, initially as a paid service offered by carriers and later as a mere capability of an Internet service or a feature of network software.)
The first of the networking services to affect today's data networks emerged during World War II. A digital circuit was used to make encrypted trans-Atlantic telephone calls between Winston Churchill and Franklin D. Roosevelt. But the earliest computer data circuits, in the 1960s, didn't use digital circuits; they used analog circuits that transmitted data at such breakneck speeds as 1.2 kilobits a second (some of today's Internet backbones are running at 2.4 gigabits a second, about two million times faster).
Switching over to digital
Throughout the 1960s and 1970s, telephone carriers transformed their own networks to digital circuits, but carriers only began selling digital circuits to corporate users in the late 1970s. The first such digital circuits, under the AT&T brand name Digital Dataphone Service (DDS), were sold at the rated speeds of 4.8Kbps and 9.6Kbps.
Although telecommunications competition did not emerge until the Modified Final Judgment that broke up AT&T in 1985, data-network professionals by the late 1970s were already conducting cost/benefit comparisons among leased-line options. Leased voice lines, known as "tie lines," had none of the frequency-limiting factors of data services, and so at times they could carry more information at a lower cost than data services. Tie lines had broader bandwidth than data circuits, but they also had relatively poor sound quality. Rated analog data circuits were provided on "groomed circuits," in most cases giving higher quality transmissions. They allowed more data to be carried because of fewer retransmissions.
Soon there were other quality considerations. For a time, there were both analog 4.8Kbps circuits and digital 4.8Kbps circuits. The latter were considerably more expensive, but the quality was higher. Many corporations didn't have the opportunity to choose, however, because digital leased-line services were unable to emulate the "multidrop" analog (hierarchical) configurations that most mainframe networks had installed.
By the mid-1980s, AT&T was carrying its analog leased-line circuits on digital circuits, improving the quality of "analog" so much that in many cases it appeared a better buy than digital. That perception faded quickly when some network specialists realized that AT&T had placed no bandwidth restrictions on its 4.8Kbps digital circuits--meaning that they were really 9.6Kbps digital circuits!
Analog leased lines were multidropped, meaning they were capable of handling many points. But today's digital leased lines run only from Point A to Point B. Connecting more points requires multiplexing technology and additional leased lines. Also, leased lines are expensive, especially if they're not in constant use.
The Internet: the first packet service
Packet services addressed these problems. The first packet service was the Internet, but in the beginning the Internet was limited to government agencies and universities. Innovative carriers created Telenet and Tymnet, now known as X.25 services. There are plenty of X.25 services left in the world. Sprint, for example, continues to run Tymnet under the name SprintNet.
Packet services are merely more efficient ways of using circuits. They often are cheaper to the end user because a carrier can aggregate packets from various users on the same circuits. For a carrier to send packets from Atlanta to Chicago requires an underlying circuit. Companies that access packet services still must do so over circuits, usually leased lines, although X.25 and some frame relay services allow dial-in connections.
Because X.25 services are charged on the basis of packets received accurately, the carriers went to great lengths to do error checking on each packet at each relay point. This turned out to reduce the total bandwidth of the network and impose high latency.
In some ways X.25 remains more mature and flexible than later technologies. Here are a few examples: X.25 networks interface between carriers, much as Internet carriers interface; users can easily dial in to X.25 services, where dial-in is just emerging with frame relay; and X.25 networks are often value-added services, meaning that the carriers will do other things besides transport packets--such as translate between protocols.
However, by the mid-1980s X.25 was considered a slow network transport protocol, so the telecommunications standards bodies created frame relay, a packet format that allows for faster transmissions because it doesn't require error checking at each switch. X.25 is no longer a popular service in the United States, while frame relay and ATM are soaring.
ATM is one of the more misleading names in networking. Many people are surprised to hear that it's not asynchronous. The name comes from its ability to carry packet data, which is asynchronous because packets vary in length. All ATM transmissions are in the form of 53-byte cells, meaning that switches can determine network timing from the cells and the specified intervals between them. This is the definition of synchronous communications.
To carry both voice and data, and potentially such exotic applications as two-way television, ATM had to be synchronous. Unfortunately, like ISDN--a complex circuit-switching capability--ATM offers too many capabilities to be used in all the ways planned.
The Internet turns 30
Everyone tends to think of the Internet as new. Actually, it will be 30 years old in 1999. For some reason, telecommunications carriers regard the Internet as a very simple network. The Internet was designed to be so smart that it could continue to function even without any humans left alive.
Like the digital circuit, the Internet--originally called ARPANET--was born out of war, in this case the Cold War. It was, in a sense, the original doomsday machine, since it was designed to ensure that nothing could prevent the command "shoot the missile" from getting through. Over the years, the Internet has been much changed from its original technology (the original idea was that no packets could be dropped; now packet drop rates of 10% are normal). But it's still the same self-routing network as ARPANET.
The big change came in 1995, when the Internet was thrown open to commercial users. Since then the aggregate bandwidth of all Internet carriers is thought to have increased by about 10 times. The Internet is largely a Web-surfing phenomenon these days, with some 58 million North American adults regularly on line, according to a CommerceNet/Nielsen Media Research study in late 1998. This indicates a growth rate of about 2.5% a month, or 30% annually. Outside North America, there are thought to be around 100 million regular Internet users, a number that is increasing at a rate of nearly 100% per year.
The downside of virtual private networks
Virtual private networks (VPNs)--data networks that use the Internet and encryption--can be less expensive than other data services. But there's a downside. VPNs are also:
...slower in terms of throughput and latency (until a new standard is commonly used, encrypted LANs cannot use the data-compression features of modems);
...less robust in general (although AT&T's frame relay collapse shows that disasters can happen to any data service); and
...vulnerable to hackers and other security problems, hence requiring expenditure for security equipment and security staff.