Ethernet

From Hill2dot0
Jump to: navigation, search

In 1972, a team at the Xerox Palo Alto Research Center (PARC) was given the task of designing the “office of the future,” in which information would flow effortlessly from machine to machine and, hence, from user to user. Two members of that team, Robert M. Metcalfe and D.L. Boggs, first outlined the idea for broadcasting messages over a local bus network, and attaching headers to them for addressing and other control functions. Thus Ethernet was born.

That straightforward idea was perhaps a bit ahead of its time, but by the time the PC revolution began to take over the desktops of the world in the 1990s, Ethernet (since standardized as IEEE 802.3) was ready to provide simple, cheap, and fast networking to the millions of users who demanded it. Only IBM’s insistence on the promotion of their own token ring standard kept Ethernet from being the be-all and end-all of LANs at the Data Link Layer and Physical Layer.

Ethernet has come a long way from its origins in the laboratories of Xerox PARC. A thorough discussion of Ethernet is virtually impossible to have without having at least a high-level view of the major phases of this evolutionary process. To understand this evolution, a few terms must first be introduced. We can refer to these as the various domains of an Ethernet. Although the concepts may appear somewhat arcane and academic at first, as we trace the changes in these domains through the evolution of Ethernet, their significance will become clear.

Three domains of Ethernet

Wire Domain

This is the domain where all devices connected together share the same wire segment. The original Ethernet implementations using coaxial cable would have a number of end stations connected together with a single coaxial cable. All of these end stations would then belong to a single wire domain. As Ethernet implementations progressed, unshielded twisted pair (UTP) became the norm and a single end station would be connected back to a hub using a UTP cable. From that point on all wiring domains only had a single end station.

Collision Domain

The media access control scheme of Ethernet is CSMA/CD which is a contention scheme. Any transmitting end station which has a possibility of colliding with another end station transmitting at the same time is considered to be in the same collision domain. Within a collision domain only one end station can be transmitting at any one time which means all end stations within a collision domain are sharing a “pool” of bandwidth.

Broadcast Domain

An Ethernet frame has information in its header address which indicates both the intended destination of the frame as well as the source of the frame. In a unicast situation (i.e., one station sending to another) these addresses uniquely identify the stations. However it is possible for a single station to attempt to send to all stations. This is called a broadcast. This type of transmission occurs often when stations are trying to identify a device they are attempting to communicate with. Any station that can receive a broadcast transmission from another is in the same broadcast domain.

Ethernet Evolution

First Generation of Ethernet

First Generation of Ethernet

When Ethernet first emerged in the 1970s, it was deployed over a coaxial cable. It was a logical bus deployed over a physical bus. Multiple systems would attach to the same cable, so it had one wire domain. If any two systems on that wire transmitted simultaneously, they would collide, so it implemented one collision domain. A frame addressed to the broadcast address would be received by all systems, so it implemented one broadcast domain.

Indeed, in this time frame, the idea of these as separate domains was not even thought of. We only began to think in terms of domains as Ethernet evolved and we began to try to describe that evolution. As is often the case, evolutionary language emerges after the evolution occurs to describe what has occurred.

Second Generation of Ethernet

Second Generation of Ethernet

By the end of the 1980s, several manufacturers were introducing hub-based Ethernet technologies using unshielded twisted pair (UTP). This was in direct response to the growing popularity of token ring, which was already hub-based and running over twisted pair. In 1990 the first IEEE standard for Ethernet over UTP in a hub configuration was approved.

Such networks now give each station a dedicated wire segment back to the hub. The failure of a cable segment can no longer affect all systems in the Ethernet, only the device at the end of its own cable segment. This network, which is still a logical bus topology but now operates over a physical star topology, implements multiple wire domains. The chief benefits are simplification of the wiring plant and a modular architecture that is less susceptible to failure.

However, if any two systems attached to the hub transmit simultaneously, they will collide. So it remains a single collision domain. It is still a shared-bandwidth environment and a single broadcast domain.

Third Generation of Ethernet

In the early 1990s, a small company called Kalpana began marketing a revolutionary (or evolutionary?) new device. It had the appearance of a hub, but it had far more intelligence. This device could read and process Ethernet frames. A hub operates only at the bit level, laboriously repeating each bit it receives. But Kalpana’s product understood the frame structure, could read MAC addresses, and could check CRC codes.

It could also learn. The device could dynamically learn the addresses of the attached devices simply by monitoring the traffic arriving on that port and scanning the source address in the frames. It would store that information locally, and use it to make intelligent decisions about what to do with subsequent frames. If it received a frame on any port addressed to an address it had learned, it could forward that frame on that port only. If the port was busy, it could buffer the frame and forward it when the port was free: a technique known as store-and-forward.

As a hub-based Ethernet, this environment clearly had multiple wire domains. The new twist was that the Kalpana product broke up the collision domain. Instead of the network acting as one large collision domain, each port was its own collision domain. The effect of this was to give each station a small private pool of bandwidth, thereby increasing the overall throughput of the network.

The network was still a single broadcast domain, however. If a station transmitted a broadcast frame, the device would replicate the frame and forward it on all ports except the one it arrived on. If any port was busy, the frame would be queued for transmission as soon as the port was free.

This was the beauty of the approach: no change was required at the end system. End systems still received all of the unicast traffic intended for them, as well as all of the broadcast traffic. They were simply protected from having to receive most unicast frames that had nothing to do with them.

Kalpana had invented the Ethernet switch, and the industry took little notice for the first couple of years. Major players like Cisco dismissed this as “toy” technology, preferring to focus on Layer 3 routed networks. They dismissed it until Kalpana began to sell these products at incredible rates, and competing products began to emerge. Then Cisco bought Kalpana and renamed the product Catalyst.

Ethernet Hardware Components

Ethernet Hardware Components

The basic components of an Ethernet LAN are many and varied. The end devices can be anything with an Ethernet interface, including other network devices (routers, other switches, access nodes) or end systems (computers, phones, peripherals). At the core of the network is the Ethernet hub or switch that provides the focus for the physical topology. Ethernet is a bus topology that is most often implemented as a physical star. An Ethernet switch can be used to implement a metropolitan-scale Ethernet network.

IEEE 802.3 Frame and Address Structure

Frame Format

As a Data Link Layer technology, Ethernet has to define a frame structure. The first fields, which are actually outside the frame proper, are the Preamble and the Starting Delimiter (SD). To understand the role of these fields it is necessary to understand a bit about how IEEE 802.3 works.

In Ethernet, the media is idle (i.e., carries no signal) when no transmission is in progress. Therefore, a mechanism is needed to allow receivers to synchronize to the transmitter before frame transmission. All 10 Mbps Ethernet LANs specify Manchester encoding—a digital, biphase encoding scheme that embeds the timing in the transmitted signal by forcing a signal transition at the midpoint of each bit. To provide for synchronization, a LAN adapter precedes its transmission with a 7 octet preamble consisting of alternating ones and zeroes, allowing the receivers to uniquely locate the middle of the bit time and synchronize. Although the other Ethernet transmission rates use different encoding schemes they also need synchronization at the start of a transmission.

Because a receiver can successfully synchronize anywhere in the first seven octets, it is necessary to signal when the synchronization signal is complete and the frame is about to begin. To do this, a starting delimiter (SD) is introduced. The SD has the same pattern as the preamble, but ends with two consecutive 1 bits. In an Ethernet, when the medium goes from an idle state to a busy state, the receiver first synchronizes on the preamble. It then monitors the alternating pattern of ones and zeroes until it detects two ones in a row. At this point, the synchronization process is complete, and the next bits to arrive will be the frame itself.

Frame Format

The first part of the frame to arrive is the destination address (DA). The DA is the MAC address of the intended recipient(s). It can be a unicast, multicast, or the broadcast address. The source address (SA) is the MAC address of the transmitting LAN adapter. The SA is always a unicast address.

The next field can have one of two meanings: a Length field, indicating the number of octets of data contained in the Data field (which can contain up to 1500 octets of data), or a Type field, indicating the nature of the payload. Because all assigned Type codes are numbers greater than 1500, there is no confusion about the use of this field. When Ethernet is used to carry IP packets, the field is used as a Type field.

Because Ethernet uses a CSMA/CD scheme, a minimum frame size is needed to ensure correct operation of the collision detection mechanism; this minimum is 64 octets. Allowing for 18 octets of header, this means the Data field should be no less than 46 octets in length. Because IEEE wants this limitation to be transparent to the upper layers, if the Length field is less than 46, a Pad field is inserted to ensure a minimum frame size. For example, if the Length field is 20, 26 octets of Pad will be inserted after the Data field. Note that if the previous field contains a Type code, then it becomes the responsibility of the upper layers to furnish at least 46 octets of data.

The final field of the IEEE 802.3 frame is a 4 octet frame check sequence (FCS) that is used to verify the integrity of a received frame. The FCS uses a 32 bit cyclic redundancy check (CRC). A LAN adapter will discard any received frame that is incorrect. The frame is considered incorrect if 1) It does not contain an integral number of octets (i.e., a few extra bits!); 2) The FCS is incorrect; 3) The length of the Data field is not consistent with the value of the Length field; and 4) The frame is a runt (shorter than 64 octets) or jabber (longer than the maximum allowed).

Address Structure

Address Structure

Communication across a LAN involves the transmission of a frame from one LAN adapter to one or more destination LAN adapters. For a LAN adapter to determine when a transmitted frame is destined for itself, some form of addressing is required.

An Ethernet address can represent a single LAN adapter or a group of LAN adapters. The latter provides support for multipoint addressing. To distinguish between the two classes of address, IEEE has given the first transmitted bit a special meaning. This bit, called the Individual/Group (I/G) bit, is unset (0) in addresses representing a single LAN adapter (e.g., unicast), and set (1) in addresses representing two or more LAN adapters (e.g., multicast). The extreme form of multicast is called the broadcast address and includes all LAN adapters attached to a given LAN. In IEEE Project 802, the broadcast address has all 48 of its bits set to one (1).

To prevent the manufacturers of standards-compliant LAN adapters from creating LAN adapters with identical addresses, IEEE administers the address space. When a manufacturer wants to market LAN adapters, it applies to IEEE and is assigned a pool of addresses. The assignment is made by specifying a particular prefix that must be used in all LAN adapters marketed by that manufacturer. The prefix comprises the first three octets of the MAC address and is called the Organizationally Unique Identifier (OUI). The manufacturer is free to use any possible value for the remaining three octets—a total of some sixteen million possible addresses. Manufacturers are free to apply for additional prefixes as the need dictates. For additional information on OUIs, including to whom they have been assigned, visit the IEEE website.

There is also some demand for addresses that are assigned and administered by the local network manager. To support this flexibility, IEEE has specified the second transmitted bit of the MAC address as the Universal/Local (U/L) bit. Addresses administered by IEEE have the U/L bit unset (0). If a user wishes to locally administer addresses (and has the equipment to support this capability), they must set (1) the U/L bit, thereby assuring uniqueness with any universally administered (i.e., IEEE administered) addresses.

10 Mbps Physical Layer options for IEEE 802.3

10 Mbps Ethernet and Coax

10 Mbps Ethernet and Coax

Before describing the Ethernet coaxial cable options, it is important to understand the four basic components that make up the Physical Layer of an Ethernet LAN: the media, the medium-dependent interface (MDI), the transmitter/receiver (or transceiver), and the LAN adapter. Each 10 Mbps option specifies a particular media type (e.g., coax, twisted pair, or fiber). Each media type requires a specific MDI (e.g., the 8 position modular jack for twisted pair, the BNC connector for thin coax, etc.). Attaching a computer to an Ethernet always requires a device called a transceiver, which provides the transmit/receive capability and, in a shared bandwidth Ethernet, line state detection (e.g., idle, busy, colliding). Finally, a LAN adapter is needed within the attaching system.

In 10BASE5, the transceiver was always a physically separate device. Between the transceiver and the LAN adapter was the attachment unit interface (AUI), which consisted of an AUI cable and two DB-15 connectors. The AUI was also known as the medium-independent interface (MII) because it permitted the same LAN adapter to be attached to any 10 Mbps media by simply changing out the MDI on the transceiver. The 10BASE5 medium was a thick coaxial cable and the 10BASE5 MDI was a piercing tap, sometimes called a vampire style connector. 10BASE5 permitted cable runs up to 500 meters, and the attachment of as many as 100 transceivers to a single segment of cable.

In 10BASE2, the transceiver could be a separate device or it could be implemented on the LAN adapter. If the transceiver was implemented on the LAN adapter, there was no AUI. This shift began a trend which has become the norm for all Ethernet LAN adapters. The MDI for 10BASE2 was a T-type BNC connector. The medium was thin coaxial cable, which could be extended up to 185 meters and host up to 30 transceivers. In both 10BASE5 and 10BASE2, the terminator was attached to the end of the cable plant. In 10BASE2, this was accomplished by attaching it to one side of the T-connectors, which attach the devices at the extreme ends of the cable plant.

Both of these plants were subject to the 5-4-3 rule, which placed an upper limit on the size of these networks. Simply stated, this rule required that no two stations in the network be separated by more than five maximum length cable segments interconnected by four repeaters. At most, three of those segments could have attached stations (e.g., at least two had to be links between two repeaters with nothing else on the cable).

It is worthwhile to discuss the structure of the Ethernet Physical Layer names. In the name 10BASE5, “10” represents the transmission rate in megabits per second, “base” represents the signaling strategy (in this case, digital), and “5” represents the total length (in hundreds of meters) of a segment of cable before a repeater is needed. Vestiges of this naming structure persist, but the last portion no longer represents cable lengths.

Both of these technologies are now considered obsolete. They are included here for completeness and because, though they are vintage, products remain on the market and some amount of deployment continues to exist. It is amazing how long technologies continue to be used even after we have significantly better options. We still use fire, horses, and axes!

10 Mbps Ethernet and Twisted Pair or Fiber

10 Mbps Ethernet and Twisted Pair or Fiber

Like 10BASE2, the 10BASE-T transceiver can be a separate device or it can be implemented on the LAN adapter, but it is always implemented on the LAN adapter today. When it is implemented on the LAN adapter, no AUI is present. In 10BASE-T, the MDI is an 8 pin modular connector, much like the connector specified for RJ-45 (although the wiring is different). The medium is twisted pair, either shielded or unshielded, which can be extended as far as 100 meters. The unshielded twisted pair (UTP) must be Category 3 or better.

Termination for the cable segment is within the transceiver in 10BASE-T. This implies that only two devices are attached to this cable plant (i.e., the transceivers at either end). One of these two devices is typically either a multiport repeater (i.e., a hub), or interconnection equipment (e.g., a LAN switch or router). It does remain possible, of course, to put a simple point-to-point 10BASE-T cable between two computers, however, as long as the transmit/receive issues are resolved by using a rolled or cross-over cable.

10BASE F is identical to 10BASE-TRJ-45 in its physical organization, but the MDI is an optical coupler and the medium is multimode fiber (MMF). As with 10BASE-T, the transceiver can be internal or external and only two transceivers are permitted per cable run, again resulting in a star-wired configuration when a hub is located at one end of the wire segment. Distance limitations are normally 1000 meters, but can be as much as 2000 meters for certain configurations. The standard specifies three types of fiber links: 10BASE-FP, 10BASE-FB, and 10BASE-FL. FP relates to a passive star system and FB relates to repeater links where multiple repeaters are cascaded. FL is the more common, and relates to station-to-repeater or repeater-to-repeater links.

Note that the 5-4-3 rule applies here as well, but the 3 has become somewhat irrelevant in that all cable segments are now link segments with only two attached devices. However, two stations could be separated by at most four hubs, placing an upper limit on the total size of the network: about 500 meters end-to-end if it was all twisted pair, about 10 kilometers for some version of 10BASE-F, and somewhere in between for mixed network environments.

Unlike 10BASE5 and 10BASE2, 10BASE-T and 10BASE-F are alive and well. The hub is usually replaced with a switch, eliminating the 5-4-3 rule completely. This is because each port on a switch, unlike a hub, has a fully functional LAN adapter and the 5-4-3 rule applies from LAN adapter to LAN adapter. The 10 Mbps LAN adapter has essentially disappeared, however, in favor of 10/100 or 10/100/1000 Mbps autodetecting adapters. These adapters can operate as 10BASE-T adapters when they detect the device at the other end of the wire is only capable of 10BASE-T operation, but they can also operate at higher speeds if they detect a faster and more capable adapter on the wire.

10 Mbps Ethernet Media Specifications

Commonly implemented 10 Mbps Physical Layer options

The visual summarizes the characteristics of the more commonly implemented 10 Mbps Physical Layer options for IEEE 802.3.

See Also

PodSnacks

<mp3>http://podcast.hill-vt.com/podsnacks/2007q1/ethernet.mp3%7Cdownload</mp3> | Ethernet
<mp3>http://podcast.hill-vt.com/podsnacks/2007q1/csma-cd.mp3%7Cdownload</mp3> | CSMA/CD
<mp3>http://podcast.hill-vt.com/podsnacks/2007q1/ethernet_addressing.mp3%7Cdownload</mp3> | Ethernet addressing