With its ability to move large blocks of data quickly, ATM brings networking nirvana a step closer
Peter Wayner
For the last several years, the networking world has been promising the average PC user that networks sporting ATM (Asynchronous Transfer Mode) would provide the tantalizing option of ``bandwidth on demand.'' But no matter how often each of us said ``Please,'' it was still hard to find and buy such systems. This is quickly changing. The manufacturers of routing switches and NICs (network interface cards) are rolling out product after product. The major providers of long-distance network infrastructure--including AT&T, MCI, and RBOCs (regional Bell operating companies) like Ameritech--already have limited ATM networks operating or in the final stages of testing. Clearly large segments of the
network world believe that ATM is the Emerald City--and they need to get there fast.
But the journey to a fully functional, ATM-based network that links computers throughout the country will take years to complete. Although the destination sounds clear and most of the standards are set, many details are sure to change as companies define in practice the capabilities they will actually build into the ATM infrastructure. The service providers, hardware manufacturers, and users (e.g., LAN operators, WAN [wide-area network] operators, you, and I) are dancing a complicated jig, and no one is taking the lead. Vendors are trying to judge how much bandwidth to offer and how many features will be in demand, while users are trying to figure out just how much they need to buy.
In the end, the interaction between the network-service providers and users will define the overall shape and cost structure of the ATM-based network. Although the same basic standard protocols will unify the entire network, there wi
ll still be significant differences in what will be available. Some service providers will invest heavily in an approach that might be more suited to, say, transmitting video, and this will let them offer that service relatively cheaply. Others will invest in another approach that might allow them to price asynchronous data, such as E-mail, at a bargain-basement price. Everyone is guessing about the right mix of product and services and hoping that their hardware will match the market's future needs.
The wide-open nature of the market, coupled with the regionalized structure of the phone business, means that you may well see a fragmented market that will seem strange to everyone used to the modern homogeneity that has been brought by worldwide corporations such as Coca-Cola. Some services may end up being less expensive in Los Angeles than in New York, simply because the local phone companies make different choices when they install their switches. These kinds of differences on the information highway
will make life, at the very least, exciting for network managers trying to optimize their decisions. However, the differences are bound to smooth out after a few years, as everyone begins to get firm ideas on which ATM services will sell successfully.
The companies that manufacture the switches and routers used by LANs and WANs are hedging their bets. They stress the flexibility of their machines and use seductive terms such as tariff arbitrage to imply that users will be able to rework their network schemes to use the least expensive service available. This flexibility is one of the best features of ATM, even though users may grow to hate the array of choices that this freedom brings.
The bottom line is that users will need to pay attention to the details and make many choices about bandwidth, types of service, and other protocols. The one real advantage of ATM is that all the modern network-interface devices are driven by software, and it will often be easy to switch protocols just by switchin
g software.
At the Desktop Level
For the time being, most of the action will be confined to networks installed in single offices and on corporate campuses. The switches, LAN converters, and interface cards that you need to set up a local ATM network are available today. You may have trouble finding software that would make it possible to use all the neat features, such as real-time video, but this is certain to change by the middle of 1995.
LAN administrators will have two choices: They can retrofit old networks with clever hardware that links the old LAN to ATM clients outside the LAN, or they can build new networks. New ATM LANs let users manage large amounts of bandwidth without overloading the network, but it costs much more for new cabling, interface cards, switches, and software. Retrofitting is cheaper and easier, because you can reuse old cable, cards, and software, but it does not provide much of the superior performance of ATM. It allows interoperability, but performance is limit
ed by the old LAN.
The first groups to adopt local ATM networks will be small workgroups that need to move large blocks of data. Many ATM network manufacturers cite Wall Street traders and seismic prospectors at oil companies as prototypical examples. These workers need to move large amounts of data quickly. The oil companies, for instance, like to move large numerical simulations of drill sites to graphics workstations, where they can visualize the data. ATM networks are among the few solutions that offer this much raw bandwidth.
The prices for small networks like these are dropping. A basic ATM network linking about 15 workstations costs about $1500 to $2000 per desktop machine for the interface card, and about $20,000 to $25,000 for the central switch. Many manufacturers are already in this market (see the text box ``Selected Sources for ATM Products''). This network would run over C5 twisted-pair wires and deliver 155-Mbps bandwidth to each machine. Servers would probably be allocated even m
ore bandwidth because of their naturally higher demand.
Prices vary widely depending on the need for additional features and perhaps for different interface electronics. One significant difference between interface boards is in the amount of on-board buffering they provide. Under ATM, the most time-consuming task is breaking up every transmission into standard 53-byte packets and then reassembling them on the other end. Some interface cards rely on the host processor to do this work and save money. This savings, though, can be a false economy, because heavy network traffic can easily bog down a workstation by requiring virtually all its CPU time.
While high-end users will always drive the edge of new technology, the primary market will be low-end users. One company, First Virtual (Santa Clara, CA), is aiming to provide the world with 50-Mbps connections at a cost of $500 each per PC. Unlike the manufacturers of the 155-Mbps networks, the company is not yet shipping a product. It says, however, t
hat production could begin in early 1995.
It is not clear at this time whether delivering this price/performance point will be successful. The low cost will attract many users; however, the networks may be too slow to handle high-bandwidth applications such as desktop video. Much of ATM's ability to knit together real-time and bursting data will go to waste. Other manufacturers say that they plan to aim for the higher 155-Mbps standard, because customers need a solid reason to move to ATM technology and wide-open performance is their selling point. They offer hints that this level of service will be available for about $700 in 1995, but for now, they refuse to announce any products.
Many manufacturers will produce variations on the ATM theme. All the cards and systems will work together, but users will need to think carefully about the details.
Retrofitting LANs
If it weren't for budget constraints, most people would be using ATM- capable LANs today. Most users, though, won't need t
he speed or features of an ATM network until functions like videoconferencing become widespread. Even if you don't need to move large blocks of data quickly, however, you still might want to connect a LAN to a larger ATM WAN. This is why companies are making Ethernet LAN routers with an ATM backdoor. Some are making entirely new routers, while others are building gateways that plug into older routers with an FDDI (Fiber Distributed Data Interface) link. Both of these let you connect your old Ethernet LAN to the ATM world without replacing your old interface cards or software.
Connecting to ATM WANs will be especially attractive to firms that have workers with different needs. Programmers, for instance, often place heavy loads on networks when they compile. Workers who use word processors, however, usually place the greatest load on the network when they open and close files. Those workers with low network demands can stay with the old network and can use an Ethernet-ATM gateway to reach the rest of the
network.
Network managers who choose this last path will gain a fair amount of flexibility. They can place an important file server in the ATM branch of the network, and the gateway will route the traffic correctly. More important, the flexibility of ATM will let the manager move the file server to different locations, including across the country. This flexibility may help protect systems against fire and other disasters.
Outgrowing LANs
ATM is not intended simply as a high-speed replacement for LANs. The strength of the technology lies in its ability to link many networks and to switch traffic among them. The same 53-byte protocol works throughout the network. This flexibility promises to let network administrators mix and match connections to adjust the throughput to meet changing needs.
For this reason, the next growth spurt in ATM networks will occur in campus environments where users can string up their own networks. The ATM Forum, a cross-industry standards group, recently a
ccepted a formal standard known as the UNI (User-Network Interface) that specifies how different switches may communicate. Several equipment vendors demonstrated the robustness of this standard by linking the switches of many manufacturers at the 1994 Interop conference.
If you run a multisite network, you will have many different options for linking the switches. You can pull your own fiber-optic cable, run copper coaxial, or string up twisted-pair wires between buildings, if they are close enough. Or you can run the link through your local phone company's network by simply buying a connection for each machine. This approach can end up cheaper if the telephone company already has fiber installed in each building. Each network manager has to evaluate the cost of running the cable and compare it to the prices the local telephone network charges for the connection. The important fact is that the basic switches and software will work with either choice.
Top-Layer Players
The greatest confusio
n and activity seems to center around the highest level, as the service providers try to determine just how much service to offer in what areas. All the major long-distance carriers (e.g., AT&T, Sprint, Wiltel, MCI, and MFSNet) are committed to offering ATM services. Several of the local RBOCs will also offer local service and some--such as Ameritech--have unregulated segments that may compete anywhere in the country.
The phone companies are struggling to anticipate demand so that they can plan how much to spend on each part of the local phone switches. At the time this article was written, users could buy T3-class ATM service (45 Mbps) as a permanent virtual circuit. This is effectively a leased line that offers a few additional features for ATM users. You get bandwidth on demand, as long as you always demand and pay for 45 Mbps per month. There are still other advantages over pure leased lines. Your ATM switches will be responsible for breaking up all the 45 Mbps of transmitted data and assigning it
to your own internal traffic. In some cases, you will also be able to send bursts of data that are larger than 45 Mbps.
By the end of this year, some companies will begin to offer T1-level service (1.5 Mbps) as well as OC3 (155 Mbps). This service will allow you to knit a 155-Mbps LAN on the East Coast with one on the West Coast without losing any speed.
The most exciting part of ATM long-distance service will be SVCs (switched virtual circuits), and these will not emerge until well into 1995 and 1996. They will provide pure nationwide bandwidth on demand, limited only by the speed of your own local service. You will pay only for the circuits as you use them. One of the barriers to the emergence of this service is the telephone companies' internal networking standard known as the BICI (Broadband Inter-Carrier Interface), which governs the way that telephone company switches exchange data and call-setup information for billing purposes.
When the BICI standard is finished, the companies wil
l have no trouble linking their services. That means that an East Coast RBOC like Bell Atlantic will be able to join with a company that services only another region of the country; together they will be able to serve customers who need connections between the two locations.
The Transition
Today, many network integrators who need to link up a WAN successfully are making interim choices that can be changed later. Many of the switch manufacturers, for instance, make switches that can spit out packets in some of the older standards, such as Frame Relay. These switches use software to handle the protocol-level bundling, which means that you can upgrade them to ATM specifications when the time is right.
Some customers may never switch over to ATM, because they find that the fee structure for Frame Relay is better suited to their needs. They may never need to use the guaranteed bandwidth feature of ATM. This approach will probably be available only for the short run if ATM becomes popular. Suppo
rting multiple standards costs money, and the most popular ones tend to drive out the others.
Paying the Pipe-er
It's still not clear how these switched virtual circuits will be priced, and the differences between the services offered by the different carriers is sure to be intriguing. This is one of the major ways that the companies will be able to differentiate their services. It is all too likely that they will come up with a welter of creative pricing plans that will make choosing a long-distance carrier for your home phone seem simple by comparison.
One option sure to emerge will be to pay for only what you demand on a pennies-per-megabit-per-second basis. In this area, there will probably be at least two major classes of services: Class A, for guaranteed continuous data (e.g., video), and Class C, for variable-rate data that doesn't need to get to the other end as smoothly. At this point, it is uncertain how carriers will structure their prices, but one estimate is that Class A data
links could cost three times as much as Class C data links. This is based on the experience of network managers who have found that they can overbook, packing three Class C users into a single Class A.
This ratio, however, is bound to change. Some local networks may find that the demands for the Class C service are different from those in the past, and this would force them to lower the amount of overbooking. Alternatively, some carriers may find themselves with excess capacity in parts of their network and choose to lower the price on Class A service to encourage consumption. Off-peak pricing and other such features will add further complexity.
Another option that may remain popular is the flat-rate bill. This seems a bit stodgy for a technology that was developed to offer unparalleled flexibility. But it promises no surprises on the monthly bill, and this will be a big advantage for some people.
Options
The basic ATM service is as close to a commodity as it can get. Bits must get
from one place to another, and that's that. The telephone companies are certain to invent as many features as they can, because they realize that highly competitive commodity markets are the enemies of profits. They are considering intriguing features. For example, Ameritech says that it will throw in Internet access with its ATM service. That means that if you happen to address your packets to an IP address, the Ameritech network will send them to the right location. The company also plans to offer off-site backup, which will allow you to store blocks of bits in their computer for a price.
Every company says that reliability standards will be an important feature. Connections with guarantees of lower downtime will cost much more than standard-rate connections. The companies will compile extensive statistics, and each user will be able to decide how to choose whether to pay extra for less downtime.
Taking the ATM Plunge
The transition to ATM promises to be exciting. The technology is prove
n. Manufacturers are shipping products. The potential applications are not written, but they are tantalizing. In essence, the table is set, the banquet is prepared, and now we need to see who comes to the party.
The speed with which people join the system will make a big difference. If everyone buys into the plan immediately, then the economies of scale will be wonderful. ATM lets you share your leased lines so well that many users may be able to get all the advantages of a leased line at a price that might be one one-thousandth of the price today. But this will happen only if the phone company can find 999 other people to share that line with you.
ATM is already likely to succeed on the corporate high end. But the market for ATM could really emerge if we see applications that engage the entire population. This service could become the equivalent of phone service in the future...if everyone wants it enough to justify the cost of laying fiber or C5 twisted-pair wires to each house in the country.
The phone companies could provide video on demand with a service like this. The possibilities for games are endless. The only thing that is uncertain is whether the general public will become interested in such a high level of service. These questions need not be addressed for several years. Until then, ATM manufacturers will have enough work to do in satisfying the anticipated demand for network users.
ATM Sources
Cascade
Communications Corp.
5 Carlisle Rd.
Westford, MA 01886
(800) 342-5926
(508) 692-2600
fax: (508) 692-9214
Chipcom Corp.
118 Turnpike Rd.
Southborough, MA 01772
(508) 460-8900
fax: (508) 624-9585
DigiBoard
6400 Flying Cloud Dr.
Eden Prairie, MN 55344
(800) 437-7241
(612) 943-9020
fax: (612) 943-5398
Fibronics
International, Inc.
33 Riverside Dr.
Pembroke, MA 02359
(800) 327-9526
(617) 826-0099
fax: (617) 826-7745
First Virtual Corp.
3393 Octavius Dr.,
Suite 102
Santa Clara, CA 95054
(408) 988-7070
fax: (408)
988-7077
Fore Systems
174 Thorn Hill Rd.
Warrendale, PA 15086
(412) 772-6600
fax: (412) 772-6500
Hughes LAN Systems
1225 Charleston Rd.
Mountain View, CA 94043
(800) 732-4252
(415) 966-7300
fax: (415) 960-3738
LightStream Corp.
1100 Technology Park Dr.
Billerica, MA 01821
(508) 262-1000
fax: (508) 262-1111
Motorola Codex
20 Cabot Blvd.
Mansfield, MA 02048
(508) 261-4000
fax: (508) 337-8004
NetEdge Systems
P.O. Box 14993
Research Triangle Park, NC 27709
(800) 638-3343
(919) 361-9000
fax: (919) 361-9060
Newbridge Networks, Inc.
593 Herndon Pkwy.
Herndon, VA 22070
(800) 343-3600
(703) 834-3600
fax: (703) 471-7080
StrataCom, Inc.
1400 Parkmoor Ave.
San Jose, CA 95126
(800) 877-0519
(408) 294-7600
fax: (408) 999-0115
SynOptics Communications, Inc.
4401 Great America Pkwy.
Santa Clara, CA 95052
(408) 988-2400
fax: (408) 988-5525
Transwitch Corp.
8 Progress Dr.
Shelton, CT 06484
(203) 929-8810
fax: (203) 926-9453
Xyplex, Inc.
330 Codman Hill Rd.
Boxborough, MA 01719
(800) 338-5316
(508) 264-9900
fax: (508) 264-9930
Illustration: Long-distance WANs can be created by hooking the local LAN ATM switches to the ATM network of the long-distance network operators.
Peter Wayner, of Baltimore, Maryland, is a consulting editor for BYTE. He can be reached on the Internet or BIX at
pwayner@bix.com
.
On The Road To ATM
