Building the Data Highway

• In the Beginning
• ATM Fixes Everything
• Why ATM?
• Routing vs. Switching
• Using ATM with LANs
• What ATM Means to You

In the beginning, there was the telephone and the global network built to carry telephone conversations around the world. Then came television, broadcasting video over the air, and, later, cable TV companies with the clever idea to broadcast video over wires. Somewhere along the way came computers and myriads of local networks built to move computer data (ethernet packets, for example) around the campus. Three types of data, three sets of wires.

Today, however, we're telecommuting, video conferencing, and distance learning. We want our "local" computer networks to span countries and move massive amounts of data from here to there without delay, and we want the same services in our home as we have at work. Separate, specialized networks won't do any more.

What we need is the data highway: a worldwide, universal-access network that can move voice, video, and computer data, and move them all equally well, whether they're going next door or around the world. 

The Internet isn't the answer for the data highway, but it is part of the answer.

What's wrong with the Internet? Its TCP/IP (Transmission Control Protocol/Internet Protocol) transport protocol suite was designed to move computer data (see Data Types and Quality of Service) over fairly unreliable networks, so the transport protocols built to work with IP (including TCP) are heavy on error checking. And, just in case something breaks along the way, IP routes each packet individually, rather than sending all of the packets from a given connection over one previously established path. This is very inefficient when you're talking about billions of packets.

Other parts of the answer will come from the telephone companies.

When you make a phone call, the telephone network's central agencies (which know exactly how to reach every phone, cell phone, pager, or whatever) work together to set up a connection that will be used for the entire call. This is a lot faster than determining the route for every single packet, and avoids the contention for bandwidth that can cripple large or busy computer networks.

I suppose it's not too surprising, then, that the major long distance telephone companies are already providing the backbone network (networks, actually, now that the Internet is being run by competing commercial interests) that the current Internet runs on.

What's wrong with the telephone network? Mostly that it was designed only to handle voice. (Voice data is easy to move; it's relatively small, comes steadily, and doesn't need to be transmitted too accurately. See "Data Types and Quality of Service" in the box below.)

Back to Contents

1. ATM is standards-based. ATM is a collaborative effort (proposed independently in the middle '80s by AT&T and several large telephone companies in Europe), designed from scratch, drawing on lessons learned in telephone and computer communication. ATM is being designed and promoted by the ATM Forum, a group of over 600 telephone companies, network equipment companies, users, and other interested parties. The work of the ATM Forum means that ATM standards are being defined by international consensus and are not dependent on the agendas of specific vendors. (But note the "is being designed" -- the ATM standards are not yet complete.)
2. ATM can simultaneously carry voice, video, and computer data. The ATM definition includes service categories that define classes of data types based on their predicted use of network services using Quality of Service and traffic parameters; see Data Types and Quality of Service. In principle, this allows the network resources on an ATM network to be used more efficiently. (Efficient use of the network is obviously good for network managers. The service categories are also good for the network users like us because we'll be able to request and pay for only the resources that we actually need.)
3. ATM can run at various speeds, over any type of physical medium. So you can build a homogenous ATM network, and still run parts of it at different speeds. In the non-ATM world, changes in network speed and/or physical media (the wires the network runs on) mean a different type of network, with different protocols, and a network bridge to convert the data from one type of network to the other.

Figure 1: Network Speeds and Protocols  (from Interphase, Corp.'s ATM Networking -- An Introduction)

 What's wrong with ATM?

Having said what's wrong with the Internet and with the global telephone network, I suppose it's only fair to ask this question about ATM. There are a number of specific things, like ATM cells are too small (and 53 isn't a power of 2; network people really like things to be a power of 2) and it doesn't support "broadcasting" (sending information without a specific destination; cable TV requires broadcasting). But what's really "wrong" with ATM is that it's not quite finished. The Quality of Service negotiations (see Data Types and Quality of Service) and how they will be applied, for example, are still being worked on.

Back to Contents

The telephone network uses circuit switching. In circuit switching, a specific path is set up for each connection, and a portion of the network's capacity is set aside for the exclusive use of that connection, whether it needs it all the time or not. This drastically cuts the network overhead; there's no contention and the work in determining the path is only done once. But it also makes less than optimum use of network capacity.

ATM networks use a hybrid between packet switching (routing) and circuit switching. ATM networks are circuit switched in that all the cells from a given connection travel on the one predetermined path, but instead of having part of the network set aside for it, the connections share the network capacity as in packet switching.

This hybrid makes ATM networks very efficient; it's not uncommon for ATM networks to operate at 90% of their theoretical capacity. And the efficiency of ATM networks increase with increasing traffic. On the other hand, a busy ethernet might run at 40% or less of its theoretical capacity, and the percentage decreases as the traffic on the network increases.

How does ATM switching work? The path used by each connection on an ATM network is assigned a unique number, a Virtual Channel Identifier or VCI. The VCI is placed in the header of each of the connection's cells. So instead of determining the path for each packet as routers do, ATM switches need only to look the cell's VCI up in its routing table to see where to send it.

But there are two questions left: how to keep an ATM network from being overused, and how to identify cells that need special handling, cells containing real-time data that should receive delivery priority, for example. For both, ATM uses service categories based on Quality of Service parameters; see Data Types and Quality of Service. This information is also carried in each cell's header. 

Figure 2: ATM Architecture  (Adapted from an ATM Forum slide.)

Like all network protocols, ATM is layered. As shown in the figure, the AAL (ATM Adaption Layer) divides data into the ATM cell's 48-byte payloads. This is only done at the end points of the ATM connection, not at ATM switches. An ATM switch might have to change the ATM cell's address header (the ATM Layer), and perhaps change the cell's physical format (the Physical Layer), but it never has to make any changes in an ATM cell's payload. This makes ATM switches much simpler (and faster and cheaper) than a network router or bridge.  Network protocols are sometimes described in terms of the OSI reference model; ATM fits into two OSI levels: level 3, network (along with with IP and IPX) and level 2, datalink (with ethernet and token ring).

IP over ATM

Classical IP over ATM, which works on all-ATM networks only: define ATM protocols for running TCP/IP over an ATM network.

LAN Emulation

ATM's LAN Emulation -- LANE -- is a much more flexible answer, one that we use here at UIC. A LANE can contain ATM and non-ATM machines or network segments. (An "ATM machine" has an ATM network interface card rather than, say, an ethernet card.) LANE client software and LANE server machines make the overall LANE look like an ethernet, giving you TCP/IP support along with ATM's speed.

Switched ethernet is another, simpler way to improve LAN performance with ATM, one that we're beginning to use here at UIC. In switched ethernet, desktop machines are attached to a specialized switch (a.k.a. switched ethernet hub). On the desktop side, the switch talks ethernet, so the desktop machines think that they are on an ethernet subnet, usually consisting of only one or a small number of machines. On the network side, the switch talks ATM and is directly connected to the ATM backbone network. This significantly increases the desktop machine's performance and security (the major security concern on an ethernet network is "sniffing" by other machines on your subnet), without requiring any changes in the desktop machine or its software.

Back to Contents

Comments are welcome; send them to
Judith Grobe Sachs, judygs@uic.edu