Getting Data from Here to There: How Computers Share Data
- Jul 16, 2004
In this hour, we will first discuss four common logical topologies, starting with the most common and ending with the most esoteric:
In the preceding hour, you read a brief definition of packet-switching and an explanation of why packet switching is so important to data networking. In this hour, you learn more about how networks pass data between computers. This process will be discussed from two separate vantage points: logical topologies, such as ethernet, token ring, and ATM; and network protocols, which we have not yet discussed.
Why is packet switching so important? Recall that it enables multiple computers to send multiple messages down a single piece of wire, a technical choice that is both efficient and an elegant solution. Packet switching is intrinsic to computer networkingwithout packet switching, no network.
In the first hour, you learned about the physical layouts of networks, such as star, bus, and ring technologies, which create the highways over which data travels. In the next hour, you learn about these topologies in more depth. But before we get to them, you have to know the rules of the road that determine how data travels over a network. In this hour, then, we'll review logical topologies.
Before discussing topologies again, let's revisit the definition of a topology. In networking terms, a topology is nothing more than the arrangement of a network. The topology can refer to the physical layout (which we discussed in Hour 1, "An Overview of Networking," and really deals with the wiring, more or less) or the logical layout of the network.
Logical topologies lay out the rules of the road for data transmission. As you already know, in data networking, only one computer can transmit on one wire segment at any given time. Life would be wonderful if computers could take turns transmitting data, but unfortunately, life isn't that simple. Computers transmitting data have all the patience of a four-year-old waiting in line at an ice-cream parlor on a hot summer day. As a result, there must be rules if the network is to avoid becoming completely anarchic.
In contrast to physical topologies, logical topologies are largely abstract. Physical topologies can be expressed through concrete pieces of equipment, such as network cards and wiring types; logical networks are essentially rules of the road that those devices use to do their job. In other words, this is software.
When packet switching was young, it didn't work very efficiently. Computers didn't know how to avoid sending data over the wire at the same time other systems were sending data, making early networking a rather ineffective technology. Just think about itit was similar to two people talking on the phone at the same time.
Ethernet, invented in 1973 by Bob Metcalfe (who went on to found 3Com, one of the most successful networking companies), was a way to circumvent the limitations of earlier networks. It was based on an IEEE (Institute of Electronic and Electrical Engineers) standard called 802.3 CSMA/CD, and it provided for ways to manage the crazy situation that occurred when many computers tried to transmit on one wire simultaneously.
The foundation of ethernet is a method of transmitting data called CSMA/CD, or Carrier Sense Multiple Access/Collision Detection. It sounds complicated, but it's quite simple; it's a protocol based on common sense. Here's how it works, in a blow-by-blow account.
In an ethernet network
All computers share a single network segment, called a collision domain. A collision domain is the group of computers that communicate on a single network wire, also called a segment. The segment is a collision domain because if there's more than one computer in it, it's a cinch that at some point those computers are going to try to transmit data simultaneously, which is a big no-no.
Each computer in a collision domain listens to all transmissions on the wire.
Each computer can transmit data only when no other computer is currently transmitting.
Each computer listens for a quiet time on the wire (this is the carrier sense multiple access) in CSMA/CD). When the network wire is quiet (which is measured in nanosecondsnetwork quiet has no relationship to human quiet), a computer that has packets of data to transmit sends them out over the network wire. If no other computers are sending, the packet will be routed on its merry way.
When two computers transmit packets at the same time, a condition called a collision occurs (this is the collision detection part of CSMA/CD). In terms of networking, a collision is the thing that happens when two computers attempt to transmit data on the same network wire at the same time. This creates a conflict; both computers sense the collision, stop transmitting, and wait a random amount of time (in nanoseconds) before retransmitting. The phrase "random amount of time" is important because it's key to reducing collisions and it's unlikely that more than one computer on a network will randomly select the same number of nanoseconds to wait until resending.
The larger the collision domain, the more likely it is that collisions will occur, which is why ethernet designers try to keep the number of computers in a segment (and hence a collision domain) as low as possible.
Take a look at Figure 3.1 to see a diagram of an ethernet topology.
If a second computer tries to transmit data over the wire at the same time as the first computer, a collision occurs. Both then cease transmitting data, wait a random amount of time for a quiet period, and transmit again; usually this solves the collision problem. It is really that simple.
CSMA/CD, as personified in ethernet, does have some problems. Sometimes a network card goes into a mode in which it fails to obey CSMA/CD and transmits all the timethis is called jabber, and it's caused either by faulty software or a defective network card. Other problems can be caused by a segment with too many computers, which causes too many systems to try to transmit at each quiet time; this can cause broadcast storms. Fortunately, newer forms of ethernet (specifically switching, which we'll discuss further on) can circumvent these limitations by segmenting the network very tightly.
FIGURE 3.1 An ethernet topology: Only one computer can transmit data at a time.
Ethernet's Nuclear Family
Ethernet is broadly used to describe both the logical topology that uses CSMA/CD and the physical topologies on which CSMA/CD networks run. All the basic ethernet topologies are described in IEEE standard 802.3. The members of the nuclear family are listed here:
10BASE-2, or coaxial networking. The maximum segment length of 10BASE-2 is 185 meters. This is a dead technology and is not used for new installations.
10BASE5, or thicknet. Thicknet is also called AUI, short for Attachment User Interface. AUI networks are an intermediate step between 10BASE-2 and 10BASE-T. 10BASE5 is a bus interface with slightly more redundancy than 10BASE-2. The maximum length of a 10BASE5 segment is 500 meters. Like 10BASE-2, this is a dead technology and is not used for new installations.
10BASE-T, which runs over two of the four pairs of unshielded twisted-pair wire. In 10BASE-T, the maximum cable length from the hub to a workstation is 100 meters. 10BASE-T is pretty much dead technology at this point.
Fortunately, the ethernet standard has grown to include faster networks and fiber-optic media. The newer members of the ethernet family are described in IEEE Standard 802.3u, and include these:
100BASE-T, also called fast ethernet, in which data travels at 100 megabits per second over two pairs of unshielded twisted-pair copper wire. The maximum cable length between the concentrator and the workstation for fast ethernet is 20 meters, and it requires Category 5 cabling standards. (Cable standards will be discussed later.)
100BASE-FX and 100-Base-FL, which is fast ethernet running on optical fibers. Because optical fibers can carry data much further than copper wire, 100BASE-FX and FL have much greater maximum cable lengths than 100BASE-T.
1000BASE-T, also called gigabit ethernet, allows data to travel at one gigabit (1000 megabits, or ten times faster than 100BASE-T) per second. Currently, 1000BASE-T is used mostly for servers and for organizational backbone networks, but over time that will surely change. By the next edition of this book, it will probably be common to have gigabit ethernet to the desktop. This topology runs on CAT 5E or CAT 6 copper wire and over fiber.
Token Ring and FDDI
Ethernet CSMA/CD networks provide a relatively simple way of passing data. However, many industry observers correctly note that CSMA/CD breaks down under the pressure exerted by many computers on a network segment. These observers are correct; the squabbling and contention for bandwidth that is part and parcel of ethernet does not always scale efficiently.
In an attempt to circumvent this problem , IBM and the IEEE created another networking standard called 802.5. (Does anyone see a pattern here? Every new invention is built to rectify the older standard's shortcomings.) IEEE 802.5 is more commonly identified with token ring; FDDI also uses the 802.5 method of moving data around networks.
Token ring works very differently from ethernet. In ethernet, any computer on a given network segment can transmit until it senses a collision with another computer. In token ring and FDDI networks, by contrast, a single special packet called a token is generated when the network started up and is passed around the network. When a computer has data to transmit, it waits until the token is available. The computer then takes control of the token and transmits a data packet. When it's done, it releases the token to the network. Then the next computer grabs the token if it has data to transmit (see Figure 3.2).
FIGURE 3.2 A token ring topology (FDDI works in the same fashion): The only computer that can transmit is the computer holding the token.
In comparison to the contentious nature of ethernet, token ring and FDDI appear quite civilized. These two logical topologies do not have collisions in which multiple stations try to send data; instead, every computer waits its turn.
Token ring's underlying methodology worked to my advantage while a teaching assistant in grad school. To reduce my tuition, I taught a couple of sections of freshman composition, and one of the sections was particularly fractiousstudents would interrupt each other willy-nilly, which got in the way of teaching and learning. So I brought in a koosh ball (one of those little rubber hairy plastic things that were all the rage in the early 1990s) and announced that only the student with the koosh could talk. It took a couple of classes for the rules to sink in, but once they did, the class was much more civiland a better place for learning. Who says high technology has no place in the real world?
Token ring suffers slightly fewer bandwidth-contention issues than ethernet; it holds up under load fairly well, although it too can be slowed down if too many computers need to transmit data at the same time.
Despite its load-friendly architecture, IBM held token ring as a proprietary topology for too many years. By the time it allowed other manufacturers to begin making devices for token ring, the battle for market share was already lost to lower-cost ethernet devices. It also never really evolved to transmit data as fast as ethernet, which might have sealed its fate. Currently, token ring appears to be dying out one network at a time, and I haven't installed a new token ring device in about four years.
Asynchronous Transfer Mode (ATM)
ATM networking is the youngest of the topologies described here. It was designed to circumvent the shortcomings of existing topologies, and hence it was created from whole cloth. Unlike ethernet, token ring, or FDDI, it can natively carry both voice and data over network wire or fiber. ATM transmits all packets as 53-byte cells that have a variety of identifiers on them to determine such things as Quality of Service.
Quality of Service in packet data is very similar to quality of service in regular mail. In regular mail, you have a choice of services: first class, second class, third class, bulk mail, overnight, and so forth. When you send an overnight message, it receives priority over first-class mail, so it gets to its destination first.
A few bits of data in a packet of data indicate the quality of service required for that data. When the Quality of Service feature is implementedas it is in ATM and Internet Protocol version 6 (IPv6)you can send packets based on their need for bandwidth. For example, email is relatively low priority and might be given third-class service; video or audio content, which has to run constantly, gets a higher priority.
ATM is fast. At its slowest, it runs at 25 megabits per second; at its fastest, it can run up to 1.5 gigabits per second (which is why phone companies use it for some of the huge trunk lines that carry data for long distances). In addition to its speed, ATM is exponentially more complex than either ethernet or token ring. Most commonly, the 155 megabit per second speed of ATM is used for applications in which quality of service and extraordinary speed are required.
Currently, ATM equipment is both esoteric and expensive. Fore Systems and IBM have both invested heavily in ATM-to-the-desktop technology (that is, they use ATM to link servers and workstations) and are banking on the need for multimedia networks over the next several years. ATM standards and interoperability are still touch-and-go, however.
Unfortunately, ATM, like token ring, has largely been eclipsed in the consumer market by Fast ethernet and Gigabit ethernet, which provide comparable performance at lower cost.
That just about wraps up the discussion of logical topologies. Now it's time to discuss protocols.