Ethernet Overview

Ethernet is a physical and Data Link layer networking standard defined by the IEEE 802.3 set of protocols. These protocols define frame size, connection rules, cable types and lengths, transmission speeds, and a host of other specifications. This overview of Ethernet is intended for background only. For more detailed information on Ethernet specifications, see References at the end of this chapter.

Architecture and Theory

An Ethernet can be either a bus topology or a star topology. Bus topologies have multiple stations connected to a cable or segment. Star topologies have stations connected to a central device such as a hub (repeater) or a switch. An Ethernet repeater (or hub) connects Ethernet segments together at the physical layer. A repeater essentially extends a segment. A bridge or switch connects Ethernet segments together at the data link layer. Ethernets on either side of a bridge or switch are separate physical Ethernet segments. Ethernets on either side of a repeater are the same Ethernet. Physical segment refers to a collision domain, the extent of cabling where length and Ethernet collision timing rules apply.

Access Method

Ethernet stations can attempt to transmit any time they need to when the wire is not in use. They use a method known as Carrier Sense Multiple Access Collision Detection (CSMA/CD) to govern access to the media. An Ethernet station listens to make sure that the Ethernet is quiet (no carrier) before transmitting a frame. During transmission, the station listens for a set period to make sure that the frame did not collide with another frame. As long as the Ethernet is built with a precisely defined set of limitations, a station will always hear when a collision occurs. A collision occurs any time two or more stations attempt to transmit at the same time. If a collision occurs, each station involved uses a back off algorithm to determine when to retransmit the frame. The back off algorithm helps to ensure that no stations will try to re-transmit at the same time and thus collide again.

Collision Domain

An important concept is the collision domain. A collision domain is a system of Ethernet segments and repeaters (or hubs). A collision domain stops at the Data Link layer. A bridge (or switch) connects two or more collision domains.

Especially in older Ethernet networks, proper management of the collision domain was the most important factor in a properly functioning Ethernet. Making sure that all cable types, lengths, and repeater counts were within specification would go a long way toward keeping your Ethernet running properly. With the declining costs of switches, Ethernet networks that violate these specifications are getting rarer as collision domains shrink. But even in switched Ethernet networks particularly 100 Mbps or higher it is very important to follow cable and length restrictions.

An Ethernet repeater (or hub) is used to extend an Ethernet, but there are limitations. An Ethernet frame eventually runs out of either signal or time. Repeaters will reform the signal but they cannot overcome the timing issue. Therefore, there are strict limits to the number of repeaters one can use within a collision domain.

For 10-megabit Ethernets, there is the 5-4-3 rule:

• No more than 5 repeated segments

• No more than 4 repeaters between any two stations

• Only 3 of those segments may be populated with stations (the other two would be just link segments a segment connecting two repeaters with no other stations)

For 100-megabit Ethernets, there are two types of repeaters: class I or II. Class I repeaters can connect different media types (copper to fiber). Class II can only connect similar media types. Only one class I repeat can be used in a collision domain. Only two class II repeaters may be used in a collision domain. The two classes of repeaters cannot be mixed within a collision domain.

Standards

There are several types of Ethernets that vary in cable types and topology. Here are the most common standards:

• 10Base2, 10Base5 Bus topologies based on coaxial cable with multiple stations on a single segment of cable. The 10Base corresponds to a transmission speed of 10 megabits per second. The 2 and 5 correspond to the maximum segment lengths: 200 and 500 meters. Stations transmit and receive on the same cable. Neither 10Base2 nor 10Base5 is commonly deployed currently. 10Base5 is also known as thicknet; 10Base2 is also known as thinnet. These names came from the difference in cable sizes.

• 10BaseT, 10BaseFL Star topologies based on twisted pair or fiber and separate transmit and receive channels. Each station is connected to a central hub (repeater). Although transmission speed is still 10 Mbps, the separate transmit and receive channels make full-duplex operation (receiving and transmitting simultaneously) possible. 10BaseT and 10BaseFL links are either half- or full-duplex. Both stations on either end of the link must be configured the same. Almost all repeater hubs are half-duplex. The maximum length for 10BaseT is 100 meters. The maximum length for 10BaseFL is 2000 meters.

• 100BaseTX, 100BaseFX Star topologies based on twisted pair and fiber cables. Stations are connected to a central hub (or often to a switch). As in 10BaseT/FL, full-duplex operation is possible. But the transmission speed is increased an order of magnitude to 100 megabits per second. For 100BaseTX, the maximum cable length is the same as for 10Base-T, but the cable must meet Category 5 specifications. In fact, it is not only important that the cable meet category 5; the patch panels, terminations, and wall jacks must also meet category 5. The maximum length for a 100BaseFX segment is 412 meters.

• Gigabit Ethernet A star-based topology currently supporting only fiber (although copper twisted pair will be available soon). Currently only switches are available there are no gigabit Ethernet hubs. And as the name implies, the transmission speed is 1 gigabit per second (1000 Mbps).

Switched Ethernet Versus Shared Ethernet

Virtually all Ethernets built today are star topologies based on the 10/100BaseT/F standards. It is the device at the center of the star that causes controversy. In 10 M or 100 M Ethernets, the central device may be either a shared hub or a switch. With the declining price in switches, many designers opt for fully switched Ethernets, which means there are no shared segments in the network.

Concern for collision rates also has been a motivation for migrating to fully-switched networks. In fully-switched, full-duplex Ethernets, collision counts are of no concern and should be 0. However, this motivation is misguided. Although monitoring collision rates in a shared Ethernet is important, collision rates should not be primary indicators of the health of your Ethernet. Ethernet collisions are normal. Collisions provide the access control mechanism. If there were no collisions on a shared segment, it would mean that no one (or perhaps only one) was transmitting.

An often-cited bit of conventional wisdom is that collision counts should total no more that 1 percent of the frames transmitted. However, changing to a fully-switched network to reduce collision rates to 1 percent may actually degrade the performance of your network. Switches introduce more latency than a hub. So, for some applications, a shared hub may perform better than a switch. It is important to choose whatever works best for your needs and to manage it accordingly.

Collision rates of 5 percent or even 10 percent can be fine. It is more important to monitor true Ethernet errors such as CRC errors, late collisions, runts, or jabbers. (These terms are discussed in more detail in "Error/Fault Detection" later in this chapter.) The MAC layer retransmits a frame destroyed by a collision almost instantaneously. In contrast, the MAC layer discards frames corrupted by cable problems or late collisions. For these problems, it is a higher layer's responsibility to request retransmission, and then only after a lengthy timeout. The performance impact of a 10 percent collision rate pales in comparison to even a 0.1 percent CRC error rate.

Some useful guidelines for collision rates for a given utilization are shown in Table 13-1. However, these numbers are only guidelines. More accurate numbers are obtained from baselining collision counts for your network under normal operations.

Table 13-1. Collision Rates Versus Utilization

Segment Utilization	Maximum Percent Packets Colliding
0 19%	1%
20 49%	5%
>50%	15%

There are much better reasons to use Ethernet switches. Moving to switches allows full-duplex operation between stations and switches, which effectively doubles the bandwidth. The capabilities of VLANs to partition LANs and facilitate Moves, Adds, and Changes are another reason to migrate to a switched environment.