Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern..

Page 1 of 36

[Figures are not included in this sample chapter]

Switched, Fast, and Gigabit Ethernet

-3-
Ethernet, Fast Ethernet, and Gigabit Ethernet
Standards
This chapter discusses the theory and standards of the three versions of Ethernet around today:
regular 10Mbps Ethernet, 100Mbps Fast Ethernet, and 1000Mbps Gigabit Ethernet. The goal of this
chapter is to educate you as a LAN manager or IT professional about essential differences between
shared 10Mbps Ethernet and these newer technologies. This chapter focuses on aspects of Fast
Ethernet and Gigabit Ethernet that are relevant to you and doesn’t get into too much technical detail.

Read this chapter and the following two (Chapter 4, "Layer 2 Ethernet Switching," and Chapter 5,
"VLANs and Layer 3 Switching") together. This chapter focuses on the different Ethernet MAC and
PHY standards, as well as repeaters, also known as hubs. Chapter 4 examines Ethernet bridging, also
known as Layer 2 switching. Chapter 5 discusses VLANs, some basics of routing, and Layer 3
switching. These three chapters serve as a precursor to the second half of this book, namely the
hands-on implementation in Chapters 8 through 12. After you understand the key differences between
yesterday’s shared Ethernet and today’s Switched, Fast, and Gigabit Ethernet, evaluating products and
building a network with these products should be relatively straightforward.

The chapter is split into seven sections:

l "Ethernet and the OSI Reference Model" discusses the OSI Reference Model and how Ethernet
relates to the physical (PHY) and Media Access Control (MAC) layers of the OSI model.

l The second section, "10Mbps Ethernet MAC and PHY Standards," delves into the 10Mbps
Ethernet MAC in more detail and discusses the various Ethernet physical layer
implementations available today: Thick, Thin, Twisted-Pair, and Fiber Ethernet. You can skip
this section if you are very familiar with Ethernet. We included it as a refresher to set the stage
for our discussion of the new Fast Ethernet standard and switching technology.

l Ethernet was designed to be a shared-media, half-duplex technology. The third section of this
chapter, "Half- and Full-Duplex Ethernet," discusses the transition from half-duplex to full-
duplex transmission.

l The fourth section of this chapter, "100BASE-T/Fast Ethernet," intro-duces the IEEE 802.3u
Fast Ethernet standard and discusses how it differs from the regular 10Mbps Ethernet standard.
This section explains the Fast Ethernet MAC and the four different PHY standards in more
detail.

l In the section "Gigabit Ethernet," you learn about the new Gigabit Ethernet MAC and PHY
standard, as well as the new 1000BASE-T PHY for Category 5 or better UTP cable.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 2 of 36

l Next, we discuss the auto-negotiation standard. Auto-negotiation was first introduced as part of
the Fast Ethernet standard. It is an automatic speed-selection method, connecting different
devices together.

l In the final section, "Ethernet Repeaters," we start explaining deployment by means of

repeaters. In particular, this section discusses Ethernet and Fast Ethernet repeater standards, as
well as the Gigabit Ethernet buffered distributor/repeater design specification. We also cover
repeater design rules.

Chapter 6, "Cabling and More on Physical Layers," covers the Fast and Gigabit Ethernet physical
layers and the corresponding encoding and cabling specifications in much more detail.

Ethernet and the OSI Reference Model

Most data communications protocols in use today are defined in terms of a layered model called the
Open Systems Interconnection (OSI) Reference Model. Table 3.1 describes the layers of this model.

TABLE 3.1 NOVELL'S NETWARE OPERATING SYSTEM ADHERES TO THE ISO/OSI

REFERENCE MODEL MORE CLOSELY THAN MOST OTHER OPERATING SYSTEMS

Layer Layer Examples Comment

Name Number
Application 7 NetWare NLMs Provides access to the network.
Presentation 6 Formats data for workstation-based
interpretation/display.
Session 5 NetWare Core Protocol (NCP) Layers 5-7 are often not clearly defined and
vary by operating system. Some
applications span all three layers.
Transport 4 SPX (Sequenced Packet Provides end-to-end management, including
Exchange) acknowledgment, error correction.
Network 3 IPX (Internetwork Packet Routes packets. Dissimilar LANs
Exchange), IPX RIP (IPX communicate through Layer 3 routing.
Routing Information Protocol)
Data link 2 LLC: IEEE 802.2 LLC Bridges are Layer 2 devices. Generates
standard source and destination addresses. AUI
interfaces between Layers 1 and 2.
MAC: IEEE 802.3 Ethernet,
Fast Ethernet, Gigabit
Ethernet
Physical 1 UTP, coax, fiber Specifies the electrical and signal/coding
characteristics on the cable. Repeaters
operate at this level.

Both ANSI and the IEEE have used this seven-layer model in the past for good reason. Breaking
down a technology into different layers allows a given layer to be changed without impacting the

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 3 of 36

remainder of the model. The IEEE was able to add unshielded twisted-pair support to Ethernet while
still keeping Ethernet’s core intact--just as different software protocols, such as IPX, TCP/IP, and
NetBEUI, can be used with the same hardware because each component forms an independent layer.
In this way, interoperability between network applications is greatly improved. Next, we discuss
Layers 1 through 3 in more detail. (Layers 4 through 7 are not relevant to this book, so we do not
cover them here.)

Layer 1: The Physical Layer

The physical layer, or PHY, protocol defines the electrical signaling, symbols, line states, clocking
requirements, encoding of data, and connectors for data transmission. An example of a PHY layer is
10BASE-T, which uses Manchester encoding to transmit data. Repeaters are Layer 1 devices in that
they only retransmit signals without decoding them. That means a repeater has no MAC, and the data
is only handled by a PHY on the receiving and transmitting ports.

All higher layers talk to the physical layer through a predefined interface. For 10Mbps Ethernet, this
is the attachment unit interface (AUI). You can use a DB-15 connector to connect Layer 1 to Layer 2.
100Mbps Ethernet calls this interface the media-independent interface (MII), whereas Gigabit
Ethernet calls it the Gigabit media-independent interface (GMII).

Layer 1 interfaces to the actual cable by means of the media-dependent interface (MDI). The MDI for
10BASE-T is the RJ-45 connector, for example.

Layer 2: The Data Link Layer

The data link layer actually consists of two separate pieces: MAC and the Logical Link Control
(LLC). The LLC is a standardized interface between a hardware-specific MAC and Layer 3.

The IEEE has published a rather complex LLC specification, called 802.2. All IEEE-approved
MACs, such as Ethernet and Token Ring, use this LLC interface to talk to Layer 3. Even some non-
IEEE defined MACs, such as FDDI, chose to use this LLC as a standardized, independent interface.
The LLC function happens primarily in software.

The MAC layer is of particular interest in this book. The MAC describes how a station schedules,
transmits, and receives data in a shared-media environment. All Layer 2 MACs send and receive
frames. The MAC generates the physical source and destination addresses for a particular frame,
ensures reliable transfer of information across the link, synchronizes data transmission, recognizes
errors, and controls the flow of data. The IEEE has defined a number of MAC/Layer 2 specifications
over the years, with Ethernet being the most common one. Other IEEE-defined MACs are Token
Ring/802.5, Token-Bus/802.4, and 100VG-AnyLAN/802.11. In general, MACs are very important in
shared-media environments where multiple nodes can connect to the same transmission medium. The
MAC sends and receives data via Layer 1, the media or physical layer.

You can link two different LAN segments via a bridge. You can connect a Thinnet segment
(10BASE2) to a 10BASE-T segment by means of a bridge, for example. Bridging is often called a
Layer 2 function these days. A bridge contains a MAC and PHY for every port. We discuss fast
multiport bridges, better known as Layer 2 switches, in more detail in Chapter 4.

Layer 3: The Network Layer

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 4 of 36

Ethernet, the focus of this book, operates at Layer 2. With Layer 3 switches, however, you can now
buy a Layer 2 and Layer 3 device in one box, so we have added a chapter to cover this topic.

Layer 2 uses frames, and Layer 3 uses packets. A packet comprises one or more frames: It might take
more than one frame to transfer a complete packet. The network layer is responsible for setting up the
actual connection between source and destination. Often, different paths exist between two endpoints,
and Layer 3 needs to determine the best routes for getting the packet to the final destination.
Therefore, Layer 3 is synonymous with routing.

Sometimes, dissimilar networks consisting of different types of MAC standards need to be connected
together. A company may have an FDDI backbone, for example, but workgroups are connected using
Ethernet. In this case, the network layer is required to convert between the dissimilar frames present
in the Ethernet and FDDI networks. You can do this by taking the individual frames, reassembling
the original packet, and then sending out frames in the new framing format again. Layer 3, and
routing in particular, is a very complex subject; there are many books that cover this topic.

Figure 3.1 illustrates the workings of repeaters, bridges, and routers with respect to the OSI model.

FIGURE 3.1 Repeaters operate at Layer 1 of the OSI model, bridges at Layer 2, and routers at
Layer 3.

10Mbps Ethernet MAC and PHY Standards

Ethernet is strictly based on a layered OSI model. As a result, you can easily combine the Ethernet
MAC with different PHYs. This section discusses the 10Mbps Ethernet MAC standards, and the four
major baseband PHY specifications.

The Ethernet CSMA/CD MAC

The Ethernet MAC technology is called carrier-sense multiple access with collision detection
(CSMA/CD). Figure 3.2 illustrates the CSMA/CD flow.

FIGURE 3.2 A flow diagram illustrating the CSMA/CD MAC.

CSMA/CD is the shorthand version for about seven different steps that make up an Ethernet
transmission. Notice the analogy to human speech among multiple individuals:

1. A station wanting to transmit a frame of information has to ensure that no other nodes or
stations are currently using the shared media, so the station listens to the channel first. (This is
the carrier sense part, also known as "listen before talking.")

2. If the channel is quiet for a certain minimum period of time, called the interframe gap (IFG),
the station may initiate a transmission ("talk if quiet").

3. If the channel is busy, it is monitored continuously until it becomes free for the minimum
IFG time period. At this point, transmission begins (known as multiple access, or "wait for
quiet before talking").

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 5 of 36

4. A collision (two stations transmitting on the cable at the same time) may occur when two or
more stations listen while waiting to transmit, simultaneously determine that the channel is
free, and begin transmitting at almost the same time. This event would lead to a collision and
destroy both data frames. Ethernet continuously monitors the channel during transmission to
detect collisions (collision detection, or "listen while talking").

5. If a station detects a collision during transmission, that transmission immediately stops. A

jam signal is sent to the channel to guarantee that all other stations detect the collision and
reject any corrupted data frame they may have been receiving (also part of the collision
detection, or "one talker at a time").

6. After a waiting period (called a backoff) the stations that wish to transmit attempt to make a
new transmission. A special random backoff algorithm (called binary exponential backoff, or
BEB) determines a delay time that the different stations will have to wait before attempting to
send their data again. Of course, another collision could occur after the first one, especially
when many nodes are trying to obtain access at the same time. After 16 consecutive collisions
for a given transmission attempt, the packet will be dropped. This can and does happen if the
Ethernet channel is overutilized. This is also part of the multiple access method.

7. The sequence returns to step 1.

Ethernet uses frames of data to transmit the actual information, also known as payload, from source
to destination. Like most other LANs in existence today, Ethernet transmits a frame of variable
length. The length of the frame changes because the payload or data field can vary.

The original Ethernet frame specified by Digital, Intel, and Xerox is known as the DEC-Intel-Xerox
Ethernet V2.0 frame, or just DIX or Ethernet II frame. The official IEEE Ethernet frame subsequently
replaced it. The only difference is in the 2-byte frame type/length field. Figure 3.3 shows both frame
types.

FIGURE 3.3 The official IEEE and the older DIX Ethernet II frame differ in some respects.

Let’s look at the individual fields that comprise the Ethernet frame:

l The preamble is sent to allow the receiver to synchronize with the incoming transmission and
locate the start of the frame. The preamble is a sequence of 01010101..., 7 bytes long.

l The start of frame delimiter (SFD) indicates that the MAC frame is about to commence. The
SFD octet or byte is specified as 10101011.

l The source address denotes the sender. Each node has a unique address. The first three bytes of
the address are called the block ID or organizationally unique identifier (OUI) and identify the
manufacturer of the equipment. The IEEE assigns them. Intel, for example, is identified by the
00-AA-00 (hex) address, 3Com uses the 00-20-AF address, and Cisco uses 00-00-0C. The
other three bytes are called the device ID and are assigned by each manufacturer. These are
always unique. Three bytes of device ID allows for 16 million different and unique addresses:
2^24-1 or 16,777,215 to be exact. Some of the major Ethernet vendors have shipped more than
16 million Ethernet MAC devices and have started using new OUIs. Some of the references

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 6 of 36

given in Appendix C, "Useful Web Links," show the complete list of vendor OUIs.

l The destination address specifies where the frame will be sent.

l The protocol type is a field in the original DIX V2.0 frame type. It specified which kind of
Layer 3 protocol the data contained. A hexadecimal value of 08 00 indicates a TCP/IP packet,
for example, and 81 37 indicates a Novell NetWare packet.

l The newer IEEE frame differs from the Ethernet II or DIX frame type in one key area. The
IEEE frame is much more popular these days and no longer uses the protocol type field, where
it has been replaced with the length field. This specifies the total length of the data that will be
transmitted, which can vary from 0 to 1500 bytes. If the contents of this field are 0 to 1500
bytes, you can be sure that it is the length we are discussing. If the contents of the type
length/type field are greater than 1500 bytes (for example, 8,137), we are talking about the old
DIX frame type.

l The data field can vary from 0 to 1500 bytes in length. The data field is also known as the
protocol data unit (PDU). We discuss the contents of the data field in more detail later.

l If the actual data is less than a minimum length required, the MAC will add a variable pad to
maintain a minimum total frame size of 64 bytes. If the data is longer than the maximum frame
allows, Layer 3 will typically split the packet into more than one frame.

l Finally, a frame check sequence (FCS) ensures accurate transmission. The cyclical redundancy
check method (CRC) checks for invalid frames. This value is calculated from the rest of the
packet’s data and sent along in the FCS frame. The receiving station performs the same
calculations and compares its results with the FCS transmitted with the packet. If the results are
different, the packet is rejected.

l The maximum total frame size is 1518 bytes. (The frame officially starts with the source
address.)

There are a total of four different frame types. The DIX Ethernet II frame discussed previously, plus
three different versions of the IEEE frame type: the Ethernet 802.3 frame (also known as Novell Raw
LLC), the 802.2 frame, and the 802.2 with SNAP frame. All look slightly different. Figure 3.4 shows
the contents of the 1500-byte data field for the four different frame types. The following list details
the three different versions of the IEEE frame type:

l The Novell Raw frame format is used only by Novell’s NetWare operating system, and uses the
802.3 frame shell type without adding an IEEE 802.2 LLC header within the data field. In this
case, NetWare adds its own IPX information. This type of frame is called "Novell Raw"
because it encapsulates the IPX data in raw form without any 802.2 LLC information. Because
the frame doesn’t use the IEEE 802.2 LLC information, this frame type is actually Novell-
proprietary and not 802.3 compliant.

l For the 802.2 frame type, the data field contains the 802.2 LLC-embedded information. The
IEEE frame contains the protocol type information within the LLC subframe. Three fields are
located at the beginning of the data field. The DSAP field, a 1-byte SSAP field, and a 1-byte
Control field. The IEEE frame assigns Service Access Point numbers; among those currently

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 7 of 36

defined are E0 for Novell, F0 for NetBIOS, 06 for TCP/IP, and AA for the Subnetwork Access
Protocol (SNAP).

l The 802.2 with SNAP frame is almost identical to the 802.2 frame, except it supports more
than 256 protocol types by adding a 5-byte Protocol Identification field. On any SNAP packet,
both the DSAP and SSAP fields are set to AA, and the Control field is set to 03. The 5-byte
protocol field then follows.

Table 3.2 lists all the relevant Ethernet MAC frame parameters. Most 10Mbps Ethernet/802.3 MAC
parameters are listed in bit times. The Ethernet MAC is inherently scalable. All the parameters can be
measured in terms of the time taken to transmit 1 bit of data, referred to as bit times. Note that the
actual speed of Ethernet (10Mbps) is not mentioned in the MAC specification at all. This makes it
very easy to run Ethernet at different speeds.

TABLE 3.2 KEY 10MBPS ETHERNET MAC AND FRAME PARAMETERS, SPECIFIED IN
BIT TIMES

Parameter Value (Bit Times)

Time 512 bit times
MinInterFrameGap 96 bit times (or 9.6 ms)
AttemptLimit 16 (tries)
BackoffLimit 10 (exponent)
JamSize 32 bits
MaxFrameSize 12144 bits (1518 bytes)
MinFrameSize 512 bits (64 bytes)
AddressSize 48 bits

FIGURE 3.4 The data field of the four different frame Ethernet types in use.

Calculating the time to transmit 1 bit for 10Mbps Ethernet transmission becomes very easy:

For 1Mbps Ethernet/StarLAN, the frame looks exactly the same. The only thing that changes is the
bit time. For StarLAN, the bit time is 1/1MHz = 1ms, or 1000ns. Fast Ethernet works exactly the
same way: The frame is identical again, but the bit time is reduced to 1/10, or 10ns.

Ethernet PHYs

This section looks at the different PHY implementations for 10Mbps Ethernet (see Figure 3.5). There
are officially five ways to transmit 10Mbps Ethernet:

l 10BASE5 is the original thick Ethernet coaxial cable standard, dating back to the early 1970s.

l 10BASE2, also known as thin Ethernet, was added in the early 1980s and uses a thinner
coaxial cable.

l In 1990, Ethernet over unshielded twisted-pair, known as 10BASE-T, was standardized.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 8 of 36

l 10BASE-F, although less well known, is very important because it utilizes fiber cabling to
carry Ethernet over extended distances. The physical layers mentioned so far all use baseband
transmission methods, which means the entire frequency spectrum transmits the data.

l 10BROAD36 is different from all the other Ethernet PHY standards in that it uses broadband
transmission technology to transmit. This allows different channels to communicate
simultaneously on the same cable. 10BROAD36 is far less popular and no similar 100Mbps
broadband PHY exists, so we will not discuss 10BROAD36 in this book.

FIGURE 3.5 The different Ethernet/802.3 PHYs.

10BASE5: Thicknet

10BASE5 is the original Ethernet 802.3 standard. 10BASE5 utilizes a thick, often yellow, coaxial
cable that has a physical diameter of 10mm. The cable has to be terminated with a 50Ohm/1W
resistor. Up to 100 stations per segment are allowed.

10BASE5 utilizes a bus topology, as all stations are connected via one single continuous coax cable.
The maximum length of one coax segment is 500 meters, a function of the quality of coaxial cable.

Stations using a network interface card (NIC) are attached with a DB-15 connector to the short AUI
cable. The AUI cable in turn connects to a media attachment unit (MAU) that is bolted to the coax
cable by means of a device commonly called the vampire connector. The MAU contains the actual
transceiver that connects to the coaxial cable.

For proper CSMA/CD operation, one end node needs to sense that a collision has occurred at the
other end of the wire. The maximum network diameter of an Ethernet network is limited by the time
it takes for a signal to travel from one end to the other, called the propagation delay. The network
diameter for 10BASE5 is limited to 2500 meters, consisting of five 500-meter segments with four
repeaters.

10BASE5 derives its name from the MAC 10Mbps signaling rate (10), baseband transmission
(10BASE), and the maximum 500-meter distance between stations on one segment (10BASE5).

NOTE
The IEEE Ethernet naming convention works as follows. The 10 indicates the
transmission speed in Mbps. The term BASE indicates baseband transmission, as
opposed to BROAD for broadband transmission. Originally, the last variable contained
a number to indicate segment length. The 5, for example, means 500-meter segment
length, synonymous with the original thick Ethernet standard. One would think that the 2
in 10BASE2 indicates 200-meter segment length. Yet 10BASE2 is specified to be only
185 meters, or 300 meters in a point-to-point configuration without repeaters. Maybe the
2 is an arithmetic mean, or 185 meters rounded up? With newer Ethernet standards, the
IEEE chose letters to describe the physical layer rather than numbers. The T in 10BASE-
T, for example, indicates unshielded twisted-pair cabling. The T4 in 100BASE-T4
indicates four pairs of unshielded twisted-pair cabling.

The different Ethernet standards have several spelling versions. Many people use lowercase to write

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 9 of 36

the word BASE, but in IEEE terminology BASE is always capitalized. Often we see a hyphen inserted
between the word BASE and the next designator. The hyphen is only inserted for readability purposes
(for example, 10BASE-F). If the standard can be read clearly otherwise, no hyphen is used (for
example, 10BASE5).

10BASE5 uses Manchester encoding to transmit data. This encoding scheme translates a logical 1 to
a 10-bit pattern, whereas a logical 0 is sent as an 01-bit pattern.

10BASE2: Thin Ethernet

10BASE2 is similar to 10BASE5 and was invented primarily to reduce the cost and complexity of
installation of 10BASE5. The differences between 10BASE5 and 10BASE2 are as follows:

l This standard uses an RG-58 50Ohm coaxial cable that is cheaper and thinner than that used
for 10BASE5, hence the name Cheapernet or Thinnet, which is short for "Thin Ethernet."

l 10BASE2 integrates the functions of the MAU and the transceiver/AUI cable onto the NIC
itself. A BNC barrel connector replaces the AUI or DB-15 connector on the NIC.

l The inferior cable quality means reduced distance limits. The maximum length of a 10BASE2
segment has been reduced to 185 meters, as opposed to 500 meters for 10BASE5.

l 10BASE2 allows only 30 nodes per segment, versus 100 nodes for 10BASE5.

l 10BASE2 retains the four repeater/five segment rule from 10BASE5, allowing a maximum
network diameter of 5 ¥ 185m = 925m. If no repeaters are used, you can extend the maximum
length of the single segment to 300 meters.

Compared with Thick Ethernet, Thin Ethernet is much easier to install, stations are easier to add, and
most importantly, Thinnet costs significantly less. As a result, Thinnet became very popular,
effectively replacing Thick Ethernet as a workgroup cabling solution.

10BASE-T: Twisted-Pair Ethernet

In 1990, the IEEE adopted the 802.3i 10BASE-T standard, a completely new physical layer standard
for Ethernet. 10BASE-T is very different from coaxial Thick and Thin Ethernet in many respects:

l 10BASE-T utilizes two pairs of unshielded twisted-pair telephone-type cable: one pair of
wiring to transmit data, and a second pair to receive data. Eight-pin modular plugs, type RJ-45,
are used as connectors.

l Just like the other Ethernet PHY standards, 10BASE-T uses Manchester encoding, but with
predistortion of the electrical signal to allow transmission over UTP. (Predistortion means the
electrical signal uses an offset voltage and doesn't always return to 0V when idle.) The
signaling frequency is 20MHz, and UTP cable (Category 3 or better) must be used.

l 10BASE-T was the first Ethernet PHY to incorporate the Link Integrity feature that makes
installation and troubleshooting cabling problems a lot easier. The two devices at each end of
the wire send out a heartbeat pulse every 16 meters. Both hub and NIC look for this signal

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 10 of 36

when connected to each other, for example. Receiving the heartbeat signal means that a
physical connection has been established. Most 10BASE-T equipment features an LED
indicating that the link is good. LAN managers typically start troubleshooting wiring problems
by looking at the state of the link LED on both ends of the wire.

l The maximum segment length is 100 meters, which is in accordance with the EIA/TIA 568-A
wiring standard. Repeater-repeater links are also limited to a maximum of 100 meters. (We
discuss 10BASE-T wiring in greater detail in Chapter 4.)

l The topology is changed to a star, and only two nodes per segment are allowed (the station and
the repeater, or repeater-repeater, or station- station with a crossover cable).

l 10BASE-T retains the four repeater/five segment rule from 10BASE5. This means a 10BASE-
T LAN can have a maximum diameter of 500 meters.

l External MAUs are allowed, but most 10BASE-T equipment integrates the functions of the
MAU in the data terminal equipment (abbreviated as DTE, also called node) or the hub itself.
(A DTE is defined as an Ethernet node that has a Layer 2 function--that is, a NIC or a bridge or
switch. A repeater is a Layer 1 device and is not a DTE.)

From a pure cabling perspective, 10BASE-T, or Twisted-Pair Ethernet as it is also known, actually
represented a step backward. That’s because coax cabling is far superior to UTP in terms of
bandwidth. Yet 10BASE-T was a major step forward for Ethernet overall. 10BASE-T became
popular overnight because it adopted the prevalent structured cabling system developed by the
telecommunications industry. This structured cabling system mandated central repeaters and a star-
shaped, planned, and structured wiring topology. This was very different and far superior to the point-
point, no-structure, single failure point coax cable method that had previously been used by both
10BASE5 and 10BASE2.

FIGURE 3.6 The three classic Ethernet standards are 10BASE5, 10BASE2, and 10BASE-T.

10BASE-F: Fiber Ethernet

10BASE-F became an official IEEE standard in 1993, although Fiber Ethernet equipment had been
available for a number of years before that. 10BASE-F is based on the Fiber Optic Inter-Repeater
Link (FOIRL) specification of 1987, which was created to interconnect repeaters using an extended
distance fiber-optic cable link.

10BASE-F utilizes duplex fiber (two strands of multimode or single-mode glass fiber), transmitting
on one strand and receiving on the other. Multimode fiber (MMF) of 62.5/125 mm diameter is most
often used with 10BASE-F to carry infrared light from LEDs. The IEC BFOC/2.5 miniature bayonet
connectors have become the de facto standard. This device is also known as the ST connector, which
was popularized by AT&T.

The umbrella term 10BASE-F actually refers to three different sets of fiber-optic specifications,
namely 10BASE-FL, 10BASE-FP, and 10BASE-FB. The 10BASE-FL (L, as in link) standard
replaces the older FOIRL specifications and is backward compatible with existing FOIRL-based
equipment. 10BASE-FL is the most universal 10Mbps fiber standard and can be used to connect
DTEs, repeaters, or switches. 10BASE-FL is the most widely used portion of the 10BASE-F fiber-

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 11 of 36

optic specifications, and equipment is available from a large number of vendors. The 10BASE-F
standard also incorporates the 10BASE-FB (B as in backbone), and 10BASE-FP (P as in passive)
standards. These are extremely rare, however. Table 3.3 provides a comparison of the different
10BASE-F choices. Table 3.4 summarizes the specific 10BASE-FL distance limitations.

TABLE 3.3 THE FOUR 10BASE-F OPTIONS

10BASE-F Application Number of Distance

Version Repeaters Limitation
FOIRL Prestandard, repeaters only 2 1000 meters
10BASE-FL New standard, universal, including 4 400-2000 meters
repeaters, switches, nodes
10BASE-FB Backbone repeaters with embedded AUI 2000 meters
10BASE-FP Central repeater without AC power 1 passive STAR 500 meters or 300
meters

TABLE 3.4 10BASE-FL DISTANCE LIMITATIONS FOR DIFFERENT CONNECTIONS

Connection Type Distance

Any repeater-to-DTE fiber segment 400 meters
With four repeaters and five segments 500 meters
Any inter-repeater fiber segment 1000 meters
Without a repeater (DTE-DTE or switch-switch) 2000 meters

Table 3.5 provides a summary of the 10Mbps Ethernet/802.3 PHY standards.

TABLE 3.5 SUMMARY OF DIFFERENT 10MBPS ETHERNET/802.3 PHY STANDARDS

Parameter 10BASE5 10BASE2 10BASE-F 10BASE-T

Maximum 500m 185m Varies from400m to 2000m 100m
segment length
Topology Bus Bus Star Star
Medium 50-"omega" 50-"omega" Multimode fiber (a few vendors 100-"omega"
thick coax thin coax provide for single-mode
equipment, too) UTP
Connector NICDB15 BNC ST1 RJ-45
Medium MAU bolted to External or on External or on NIC External or on
attachment coax NIC NIC
Stations/cable 100 30 N/A 2 (NIC and
segment repeater)
Maximum 5 5 5 5 segments

1 ST is the official connector, but SC and MIC are actually more popular.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 12 of 36

Half- and Full-Duplex Ethernet

Regular Ethernet is a shared-media access method. All networks that utilize a MAC are based on a
shared-media operation (hence the term media access, which implies that the transmission media are
shared and access is negotiated and not always given).

By definition, a shared-media LAN transmission method also implies half-duplex operation. Half-
duplex means that a station is either transmitting or receiving, but not both, at the same time. That’s
because in CSMA/CD, a station has to listen to see whether the channel is available, and only if it is
can a station start transmitting. When one station is transmitting data, all others have to listen.
Therefore, it is an either-or situation for all stations on the LAN. This method of operation was very
efficient for the coaxial cable on which Ethernet grew up. Running a single coaxial cable throughout
an entire office and providing everyone an access opportunity to the cable every few milliseconds was
very efficient. Sharing the transmission media also brought with it collisions, something that has
given Ethernet a bit of a bad name. The truth is that collisions are a very effective and efficient
method of preventing overload.

Full-duplex, on the other hand, means that a station is simultaneously transmitting and receiving. In
the early 1990s, many events occurred that have made Full-Duplex Ethernet a reality:

l The introduction of 10BASE-T wiring offered the capability for separate transmit and receive
data paths. Before the arrival of 10BASE-T, coaxial cable didn’t offer this capability; only one
electrical (coax) wire made simultaneous transmission and reception impossible.

NOTE
Technically speaking, a single cable can carry separate transmit and receive data
streams in two directions. This is known as dual-duplex and requires a specially
designed physical layer. The electronic circuitry required to do this is very complex, and
typically requires a sophisticated DSP. Until recently, the cost and complexity
associated with DSP technology was considered prohibitive. 1000BASE-T, however, will
utilize simultaneous transmit and receive on four pairs of UTP. With 10BASE-T, the
existing physical layer could be kept for full-duplex transmission.

l The emergence of multiport Ethernet bridges or switches meant that the physical media were
no longer being shared by multiple users but were increasingly being used to connect two
switches or a switch and a NIC together in a point-to-point manner.

In 1992, Kalpana seized on this opportunity so that it could effectively double the speed of Ethernet
through full-duplex transmission. Kalpana started working with several other industry vendors to
establish a de facto industry standard for full-duplex Ethernet over UTP wire. Kalpana proposed the
following scheme:

l One pair of UTP wire (or one fiber strand) would be used exclusively for transmission and one
for reception of data.

l A new kind of multiport bridge or switch would be required at both ends of the wire.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 13 of 36

The Ethernet MAC algorithm would be modified as follows:

l No carrier-sense (CRS) would be required, because a cable pair would now be dedicated for
both transmission and reception.

l Similarly, no collision-detection (CDT) jam or exponential backoff are required because

collisions are no longer present: They happen in a multiple-user segment only.

Essentially, full-duplex Ethernet means running Ethernet without the CSMA/ CD MAC in operation
(refer to Figure 3.2, without steps 1, 2, 3, 5, and 6). Ethernet frames are transmitted and received
simultaneously on two pairs of UTP or fiber at any time. The only possible problem is one of data
overflow. We can deal with this issue through flow control. Kalpana proposed that artificial collisions
be generated by the receiving station if an overload condition exists.

Figure 3.7 illustrates the evolution of Ethernet from a shared-media coaxial-cable LAN with
collisions to a dedicated media full-duplex technology.

FIGURE 3.7 The evolution of Ethernet from shared-media, half-duplex to dedicated media full-
duplex.

NOTE
A two-user switched connection is a prerequisite to Full-Duplex Ethernet because full-
duplex requires a point-to-point connection with only two stations present. Note that
Switched Ethernet does not automatically imply full-duplex operation.

Full-duplex has played an increasingly important role with the introduction of Fast and Gigabit
Ethernet. In 1997, the 802.3X full-duplex/flow control standard was approved. (Chapter 4 covers this
in more detail.) Some people in the IEEE even advocated making Gigabit Ethernet a full-duplex-only
technology, which means dropping the all-too-familiar CSMA/CD MAC concept altogether. In the
end, the half-duplex followers prevailed, and CSMA/CD operation was included in the standard. It is
doubtful whether we will ever see any half-duplex shared Gigabit Ethernet equipment in the
marketplace, however.

A trend is currently in place to move from repeated half-duplex to Switched full-duplex Ethernet.
With the advent of Gigabit Ethernet in particular, it looks like CSMA/CD and collisions are heading
for extinction. What will live on is the language of Ethernet, namely the 802.3 frame format. Our
well-known Ethernet frames will be zooming along at 1, 10, 100, 1000, and someday 10000Mbps.

100BASE-T/Fast Ethernet
100BASE-T is the 100Mbps version of the classic Ethernet standard. The IEEE officially adopted the
new IEEE 802.3u Fast Ethernet/100BASE-T specification in May 1995. The salient features of
100BASE-T are as follows:

l The 100BASE-T MAC uses the original Ethernet MAC operating at 10 times the speed.

l The 100BASE-T standard is designed to include multiple physical layers. Three different

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 14 of 36

100BASE-T physical layers are part of the 802.3u standard: two for UTP and one for
multimode fiber. An additional UTP PHY was added later as the 802.3y specification.

l Like 10BASE-T and 10BASE-F, 100BASE-T requires a star-wired configuration with a central
hub.

l 100BASE-T also includes a specification for an MII, a 100Mbps version of today’s AUI. The
MII layer is a digital interface connecting MAC and PHY and allows for external transceivers.

The differences between 10BASE-T and 100BASE-T are in the PHY standards and network design
areas. That’s because the new IEEE 802.3u 100BASE-T specification contains many new rules for
repeaters and network topology. Figure 3.8 provides an overview of the new IEEE 802.3u standard.

FIGURE 3.8 An overview of the 100BASE-T 802.3u standard showing MAC, MII, and the three
original PHYs. (100BASE-T2 was added as 802.3y a year later.)

The Fast Ethernet CSMA/CD MAC

The 100BASE-T MAC is almost identical to the 10Mbps "classic" Ethernet MAC. As mentioned
earlier, the 802.3 CSMA/CD MAC is inherently scalable, which means that it can run at different
speeds and be interfaced to different physical layers. StarLAN/1BASE5 took advantage of this
scalability to run Ethernet at 1Mbps. Table 3.6 compares 10Mbps and the new 100Mbps Ethernet
MAC standards. Note that the 100Mbps Ethernet MAC retains all the 10Mbps Ethernet MAC
parameters except for InterFrameGap, which has been decreased to one-tenth its original value, from
9.6ms to 0.96ms. A Fast Ethernet frame has the same framing format as a 10Mbps Ethernet frame,
except that it is transmitted across the wire at 10 times the speed.

TABLE 3.6 10MBPS COMPARED TO 100MBPS ETHERNET MAC PARAMETERS (MOST

VALUES ARE IDENTICAL WHEN MEASURED IN BIT TIMES, BUT ONE-TENTH THE
VALUE WHEN MEASURED IN REAL TIME)

Parameter Ethernet/802.3 Fast Ethernet/802.3u

slotTime 512 bit times Same1
Minimum 96 bit times (9.6ms) Same (=0.96ms)
InterFrameGap
attemptLimit 16 (tries) Same
backoffLimit 10 (exponential Same
number)
jamSize 32 bits Same
maxFrameSize 1518 bytes Same
MinFrameSize 64 bytes (512 bits) Same
AddressSize 48 bits Same

1 Same means same value in bit times as 10 Ethernet (for example, the value in the column to the
left).

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 15 of 36

You can calculate the 100BASE-T bit time as follows:

Fast Ethernet PHYs

As with 10BASE-T, 100BASE-T combines the CSMA/CD MAC with different physical layer
specifications. The IEEE 802.3u specification contains three new physical layers for 100Mbps
Ethernet:

l 100BASE-TX, which requires two pairs of Category 5 UTP or Type 1 STP cabling

l 100BASE-FX, which uses two strands of multimode fiber

l 100BASE-T4, which requires four pairs of Category 3 or better cabling

Then there is the IEEE 802.3y 100BASE-T2 PHY standard, which was completed about a year after
the original Fast Ethernet standard. 100BASE-T2 utilizes only two pairs of Category 3 cable, but is
virtually unheard of. 100BASE-TX and 100BASE-FX are the most popular 100BASE-T physical
layers. They were based on the FDDI/CDDI physical layer, utilizing the 4B/5B encoding method.
(Chapter 6, "Cabling and More Physical Layer Details," gives more information on the 4B/5B
encoding method.) 100BASE-TX is the de facto standard for high-speed horizontal connections;
100BASE-FX is becoming increasingly popular for vertical/backbone connections. 100BASE-T4 is
rarely used, whereas we know of no 100BASE-T2 products that are shipping.

Like 10BASE-T and 10BASE-F, all 100BASE-T PHY specifications require a star configuration
with a central hub. All 100BASE-T UTP PHYs use the RJ-45 connector and specify a maximum 100-
meter cable distance.

This section provides an overview of the different 100BASE-T physical layers.

100BASE-FX: Fast Ethernet for Fiber-Optic Cabling

100BASE-FX (like 100BASE-TX) borrows its physical layer from the ANSI X3T9.5 FDDI physical
layer media-dependent (fiber PMD) standard. 100BASE-FX utilizes two strands of the multimode
(62.5/125mm) fiber cabling made popular by FDDI. The maximum segment length for fiber-optic
connections varies. For two switches or a switch-adapter connection using multimode fiber, 412
meters is allowed. You can increase this number to 2000 meters if the link is full duplex. (The IEEE
did not specify SMF as part of the 802.3u standard, but numerous vendors are selling SMF-based
equipment with distances ranging from 10km to 20km.) 100BASE-FX repeater diameters can be up
to 320 meters.

100BASE-TX: Fast Ethernet for Category 5 UTP

100BASE-TX is based on the ANSI-developed copper FDDI twisted-pair physical layer-dependent

sublayer technology (TP-PMD), also known as CDDI. 100BASE-TX has numerous similarities to
10BASE-T. It utilizes the same two pairs of unshielded twisted-pair cable, the same RJ-45 connector,
and the same maximum segment length of 100 meters. As in 10BASE-T, one pair of wiring transmits
data, and the second receives data.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 16 of 36

100BASE-T4: Fast Ethernet for Four-Pair Category 3 UTP

100BASE-T4 was the only new PHY developed for 100BASE-T. It was designed to cater to the huge
installed base of Category 3 voice-grade wiring. 100BASE-T4 utilizes four pairs of voice or data
grade unshielded twisted-pair Category 3 or better cable. Unlike 10BASE-T and 100BASE-TX, no
separate dedicated transmit and receive pairs are present, so full-duplex operation is not possible.
Note that all four pairs need to be terminated for 100BASE-T4 operation.

Why Did T4 Flop?

Both 100BASE-TX and 100BASE-FX were based on the FDDI/CDDI physical layer
technology. Originally, the 100Mbps MAC and these two original PHYs were
collectively known as 100BASE-X. (The X was chosen because Fast Ethernet was based
on the 100Mbs Ethernet MAC coupled to the FDDI PHY technology.) 100BASE-X is
also synonymous with prestandard products. When the IEEE standard was approved,
however, everyone quickly adopted the official designation terminology.
Borrowing from the existing ANSI FDDI technology meant that the IEEE 802.3u
standard could be finished relatively quickly. More importantly, however, 100BASE-X
products were able to get to market very quickly. FX and TX products could be built and
sold before the actual 802.3u standard was officially ratified because existing
FDDI/CDDI semiconductor and magnetics components were available to build TX or FX
NICs, hubs, and switches.
T4 was also part of the original 802.3u specification and was supposed to cater to the
huge installed base of Category 3 voice-grade wiring. Unfortunately, T4 has been a
market flop because T4 products only started shipping a year after the standard was
approved. This two-year delta gave TX products a head start. As a result, 100BASE-TX
is by far the most widely used physical layer specification for 100BASE-T today,
accounting for more than 95% of all Fast Ethernet shipments.
T4 required four pairs, whereas both 10BASE-T and 100BASE-TX only utilized two
pairs. This issue caused a lot of confusion. The original EIA/TIA 568 cabling rules
specify four pairs to be installed, which meant using two pairs for 10BASE-T and two
spare pairs. Yet some people chose to use the two spare pairs for voice or only installed
two pairs in the first place, making the cabling plant unsuitable for T4.
In some ways, T4 was also a defensive marketing effort in the controversy with 100VG-
AnyLAN, covered in Chapters 1 and 2. At the time, the 100VG advocates made a big
deal of the fact that 100BASE-TX required Category 5 cable, whereas 100VG ran on
four pairs of Category 3. The adoption of the IEEE T4 standard effectively neutralized
the 100VG argument. Later on, the 100VG people adopted a two-pair Category 5 version
as well, but 100VG had already lost the battle by that time.

100BASE-T2: Fast Ethernet for Two-Pair Category 3 UTP

100BASE-T2 is virtually unknown. It is an IEEE standard, but no products ever made it to market.
We wanted to cover it mainly for completeness and because 100BASE-T2 technology will reappear
soon with the 1000BASE-T PHY. IEEE engineers came up with 100BASE-T2 because they thought
100BASE-T4 had two shortcomings they could improve. First, T4 requires four pairs of Category 3
wire, but some installations only have two pairs present or usable. The other issue with T4 is that it
cannot do full-duplex. When 100BASE-T4 was designed, it was thought that four pairs would be
required to transfer 100Mbps. Due to advances in digital signal processing (DSP) and integrated

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 17 of 36

circuit chip technology, it has become possible to transfer 100Mbps over only two pairs of Category 3
UTP. The IEEE worked for about two years on the 100BASE-T2 standard. By the time the standard
was complete, however, 100BASE-TX already dominated the market to such an extent that no T2
products were ever built. The following are the main features of 100BASE-T2:

l Utilizes two pairs of voice or data grade unshielded twisted-pair Category 3, 4, or 5 cable.

l Uses both pairs for simultaneous transmitting and receiving on both pairs, also known as dual-
duplex.

l Uses a more complicated five-level encoding scheme called PAM5x5. (PAM stands for pulse
amplitude modulation.)

TABLE 3.7 A SUMMARY OF THE DIFFERENT 100BASE-T PHYSICAL LAYERS

PHY 10BASE-T 100BASE-TX 100BASE-T4 100BASE-FX 100BASE-T2

Parameter
Origin of CDDI TP PMD New; developed FDDI PMD (ANSI New; developed
Technology (ANSI from 1995 to X3T9.5) from 1994 to
X3T9.5) 1996 1995
IEEE standard 802.3i-1990 802.3u-1995 802.3u-1995 802.3u-1995 802.3y-1996
Encoding Manchester 4B/5B 8B/6T 4B/5B PAM5x5
Cabling UTP UTP Category UTP Category Multimode or UTP Category
required Category 5 or STP Type 3/4/5 single-mode fiber 3/4/5
3/4/5
Signal 20MHz 125MHz 25MHz 125MHz 25MHz
frequency
Number of 2 2 4 2 2
pairs required
Number of 1 1 3 1 2
transmit pairs
Distance 100m 100m 100m 150/412/2000m1 100m
Full-duplex Yes Yes No Yes Yes
capable?

1 150m for repeater-DTE, 412m for DTE-DTE, 2000m for full-duplex DTE-DTE, and 10km for
single-mode full-duplex DTE-DTE.

Table 3.7 compares 10BASE-T with the four 100BASE-T physical layer specifications. We discuss
the different physical layers and their cabling requirements in more detail in Chapter 6.

Fast Ethernet MII

MII is similar to AUI for 10Mbps Ethernet. The MII layer defines a standard electrical and
mechanical interface between the 100BASE-T MAC and the various PHY layers. This standard
interface works like AUI in the classic Ethernet world in that it allows manufacturers to build media-

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 18 of 36

or wiring-independent products, with external MAUs being used to connect to the actual physical
cabling.

The electrical signals differ between MII and AUI. AUI has a stronger signal, capable of driving 50-
meter cable lengths; the MII signals are digital logic-type signals, capable of driving 0.5 meters of
cable. MII uses a 40-pin connector, similar to the SCSI connector, although it is smaller. Figure 3.9
shows an external MII 100BASE-FX transceiver. We don’t think you will see too many of these
because most Fast Ethernet hubs and NICs already include the transceiver.

MII never became as popular as AUI. AUI became a de facto standard because Thick Ethernet
hardware wasn’t available with an onboard transceiver. That only came later, with 10BASE2 or
10BASE-T. Yet with 10BASE2 and 10BASE-T, the AUI connector provided the capability for
backward compatibility through an external transceiver. With 100BASE-T, most hardware
manufacturers decided to skip MII and offer different products that integrate the physical layer
transceiver already. NIC manufacturers, for example, often sell the same basic design in a 100BASE-
TX and in a 100BASE-T4 version. You have to choose up front which media type you need. Think of
it as a choice between Thinnet-only or 10BASE-T-only products, as opposed to combo or AUI-only
cards using external transceivers. An example of a device with an MII connector is the Intel Pro/100
Server adapter, which uses an MII connector to provide you with the flexibility of either attaching an
external fiber or a UTP PHY transceiver.

FIGURE 3.9 Fast Ethernet MII transceivers are similar to the old Ethernet AUI transceivers, but
are very rare because most Fast Ethernet hardware includes an onboard transceiver. MII uses a 40-
pin connector very similar to SCSI-2.

Gigabit Ethernet
In 1995, Fast Ethernet seemed like such an ingenious, yet simple, idea. 100BASE-T became an
overnight success, and it was only a matter of time before the frequency would be increased another
order of magnitude. Work started on Gigabit shortly after the Fast Ethernet standard had been ratified.
Three years later, on June 25, 1998, the IEEE 802.3z standard was officially adopted.

If you have read this chapter’s information on Fast Ethernet, we should just ask you to go back and
read everything with an extra zero added everywhere. Gigabit Ethernet is that simple. Some
differences arise in the physical layers, network design, and minimum frame size that we cover here
and in Chapter 6. For the most part, however, Gigabit Ethernet is just supercharged Fast Ethernet.

The IEEE 802.3z standard includes the Gigabit Ethernet MAC, as well as three physical layers that
use the 8B/10B encoding originally developed as part of the ANSI Fibre Channel technology.

Essentially, the IEEE engineers bolted the existing Fibre Channel PHY to the Ethernet MAC running
at 10 times the speed of the Fast Ethernet MAC. The 802.3z standard encompasses two fiber physical
layer standards, 1000BASE-LX and 1000BASE-SX as well as one copper PHY, 1000BASE-CX. The
copper-based PHY was also inherited from the Fibre Channel standard and was included to enable
cost-effective and quick cross-connects. A fourth PHY, called 1000BASE-T, is still under
development.

NOTE

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 19 of 36

The ANSI X3T11 committee is responsible for the Fibre Channel technology. ANSI
decided to spell Fibre with an re rather than an er to differentiate it from the fiber-optic
cable on which it runs.

The IEEE has also defined a Gigabit MII (GMII), which is similar to the Fast Ethernet MII and
connects Gigabit MAC and PHY. The GMII is only an electrical interface specification, and unlike
the Fast Ethernet MII, the specification does not include a connector. The biggest benefit of the GMII
is that it allows circuit designers to use existing Fibre Channel PHY chips and will allow for easy
redesign of existing 1000BASE-LX equipment to accommodate future PHY transceivers.

Some people in the IEEE wanted to make Gigabit Ethernet a full-duplex-only technology; others
wanted to preserve the classic CSMA/CD algorithm and the half-duplex shared operation. The reason
for preserving the CSMA/CD part of Ethernet was twofold. First, many vendors didn’t want to
redesign their Ethernet MAC chips completely. (Running it at 10 times the clock frequency didn’t
take a complete redesign.) Second, some IEEE members just wanted to preserve the 25-year heritage
of CSMA/CD itself.

Figure 3.10 shows all the different Gigabit Ethernet components.

FIGURE 3.10 A Gigabit Ethernet block diagram showing the MAC, GMII, and four different
physical layers.

100BASE-X refers to the original Fast Ethernet MAC and FX/TX PHYs. Similarly, 1000BASE-X
refers to the 1000Mbps MAC and the LX/SX/CX transceiver technology.

The Gigabit Ethernet MAC

The main points of the Gigabit Ethernet MAC are the following:

l Gigabit Ethernet uses the official 802.3 frame format, identical to that of 10Mbps and 100Mbps
Ethernet (refer to Figure 3.3).

l Like 10Mbps and 100Mbps Ethernet, Gigabit Ethernet can operate in both half- and full-duplex
mode.

l The 1000Mbps MAC uses the original Ethernet MAC operating at 10 times the speed. Running
the Ethernet MAC at Gigabit speeds has created some challenges in terms of the
implementation of CSMA/CD. To make CSMA/CD work at 1GHz, a minor modification was
required. The slot time has been increased to 512 bytes, as opposed to 64 bytes for 10 and
100Mbps Ethernet. The slot time, in essence, is the allocated time during which the complete
frame needs to be transmitted. During the slot time, the transmitter retains control of the media.
If the transmitted frame is smaller than 512 bytes, an extra carrier extension is added. The
carrier extension concept is similar to the PAD that is added to the end of the data field within
the frame. The carrier extension is added at the end of the completed frame to make it meet the
new slot time of 512 bytes.

l IEEE 802.3z specification defines both half- and full-duplex operation. When very small
frames are transmitted in a half-duplex environment, many carrier extension bits will be added.
This makes Gigabit Ethernet very inefficient because large amounts of useless carrier extension

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 20 of 36

bits are transmitted. Assume, for example, that we wanted to transmit only 64-byte minimum
size frames. The carrier extension would add 438 bytes of carrier extension to meet the spec of
a 512-byte slot time. To calculate the overall efficiency, we still need to add the interframe gap
(IFG) overhead of 12 bytes: 64/(512+12) = 12% efficiency, or 122Mbps. This is only
marginally better than 100BASE-T! Small frames are quite common, so this inefficiency for
small frame sizes needed to be addressed. The IEEE 802.3z Gigabit MAC includes a feature
called burst mode. In this case, a station may continuously transmit multiple smaller frames, up
to a maximum of 8192 bytes worth of data. The interframe intervals will also be filled with
carrier extension so that the wire never appears free to any other stations during the burst cycle.

Table 3.8 compares the 10, 100, and 1000Mbps Ethernet MAC parameters.

TABLE 3.8 A COMPARISON OF 10, 100, AND 1000MBPS ETHERNET MAC

PARAMETERS

Parameter Ethernet/802.3 Fast Gigabit Ethernet/802.3z

Ethernet/802.3u
SlotTime 512 bit times Same1 512-byte times (4096 bit
times)
Minimum 96 bit times Same Same
InterFrameGap
AttemptLimit 16 (tries) Same Same
BackoffLimit 10 (exponential Same Same
number)
JamSize 32 bits Same Same
MaxFrameSize 1518 bytes Same Same
MinFrameSize 64 bytes(512 bits) Same Same
AddressSize 48 bits Same Same

1 Samemeans same value in bit times as 10 or 100Mbps Ethernet (for example, the value in the
column to the left).

Gigabit Ethernet PHYs

The 802.3z Gigabit Ethernet standard includes three PHYs: 1000BASE-SX and LX to support fiber-
optic cable and 1000BASE-CX for shielded 150-Ohm copper cable. Gigabit Ethernet used the same
proven concept that Fast Ethernet had already pioneered: It uses an off-the-shelf PHY standard. In
this case, the ANSI X3T11 Fibre Channel PHY was chosen. Fibre Channel was a natural choice, as it
was proven, and components were readily available. The only issue was speed. The Fibre Channel
PMD runs at 1Gbaud and uses 8B/10B encoding. This translates to a data rate of only 800Mbps. The
IEEE therefore increased the Fibre Channel PHY speed to 1.25Gbaud to obtain an actual data
throughput of 1Gbps. This means the Fibre Channel components had to be retested or even
redesigned for the 25% speed increase. See Figure 3.11 for an illustration.

FIGURE 3.11 Three of the Gigabit Ethernet PHYs are based on the ANSI Fibre Channel PHY
running at 1.25Gbaud.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 21 of 36

Finally, the 802.3ab 1000BASE-T physical layer will enable Gigabit Ethernet over Category 5 UTP.
This standard isn’t complete at the time of writing this book, but should be approved in late 1998 or
early 1999.

1000BASE-SX: Gigabit Ethernet for Horizontal Fiber

1000BASE-SX is targeted at cost-sensitive, shorter backbone, or horizontal connections (S as in

"short wavelength or shorter cable runs"). 1000BASE-SX uses the same physical layer as LX and
uses affordable 850nm short-wavelength optical diodes. 1000BASE-SX uses only multimode fiber
(MMF). The distance supported varies from 220 meters to 550 meters, depending on the type of fiber
cable used.

1000BASE-LX: Ethernet for Vertical or Campus Backbones

1000BASE-LX is targeted at longer backbone and vertical connections (L as in "long wavelength or

longer runs"). LX can use either SMF or MMF. LX requires expensive 1300nm lasers. The IEEE has
specified a segment length of 5000m for LX using SMF. For LX using MMF, the distance is 550m
(full-duplex connections). Table 3.9 summarizes the different 1000BASE-SX and LX distance
combinations. The IEEE specifies the SC connector for both 1000BASE-SX and 1000BASE-LX.

TABLE 3.9 1000BASE-SX AND LX MAXIMUM LENGTH VARIES DEPENDING ON TYPE

AND QUALITY OF FIBER CABLING USED

Wave length Fiber Type Fiber Size (mm) Bandwidth Attenuation Maximum Distance1
1000BASE-SX
850 MMF 50/125mm2 400MHz/km 3.25 500m
500MHz/km 3.43 550m
62.5/125mm3 160MHz/km 2.33 220m
200MHz/km 2.53 275m
1000BASE-LX
1300 MMF 50/125mm 400/500MHz/km 2.32 550m
62.5/125mm 500MHz/km 2.32 550m
SMF 10/125mm Huge/infinite4 4.5 5000m

1 All
distances are for full-duplex. We don’t know of a single vendor building half-duplex 1000BASE-
X equipment. For the record, half-duplex distances are 316m for all 1000BASE-LX connections and
between 220m to 316m for 1000BASE-SX.

2 The maximum length depends on the type of 50mm or 62.5mm fiber used. The 400MHz/km stuff can
only accommodate 500m, but the 500MHz/km fiber is specified at 550m.

3 Again,the maximum length depends on the type of 62.5mm fiber used. The 160MHz/km stuff can
only accommodate 220m, but the 200MHz/km fiber is specified at 275m. Most of the installed base of
62.5mm fiber has a bandwidth of 160MHz/km. The lower bandwidth 160 and 400MHz/km fiber is

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 22 of 36

more prevalent in Europe than in the United States.

4 SMF has a modal bandwidth that exceeds the capabilities of today’s electronics components. For
practical purposes, it cannot be measured and is therefore infinite.

Going the Extra Mile?

The IEEE distance limitations for Gigabit Ethernet are extremely conservative. The
Gigabit Ethernet Alliance boasts on its own Web page (www.gigabit-ethernet.org/) that it
could demonstrate 1000BASE-SX running over 380 meters at a recent tradeshow (the
specification says 220 meters). Mier Communications in New Jersey tested eight
different 1000BASE-SX switches over a 1km 62.5mm MMF fiber run. All of them ran
flawlessly.
The 5km-distance limitation was also a very conservative worst-case specification. Many
1000BASE-LX vendors guarantee their products for much longer distances: 10km seems
to be the going distance. Cisco has given its product a different name, calling it
1000BASE-LH. The caveat is that a vendor will require its equipment to be present at
both ends of the wire. Some vendors, such as Extreme and nBASE, have started shipping
fiber extenders, allowing for distances of up to 80km! This makes Gigabit Ethernet a
viable metropolitan and even wide area networking technology! Imagine a Gigabit
Ethernet WAN connection!

1000BASE-CX: Gigabit Ethernet for Copper-Twinax Cabling

The third Fibre Channel-based physical layer, 1000BASE-CX (C as in "cross-connect," or

"interconnect," or "copper"), was designed for short interconnects of hubs, switches, or routers in the
wiring closet. Copper is the preferred media because it is easier and quicker to assemble copper
jumper cables than fiber. 150-W twinax cabling, similar to IBM’s original Token Ring cabling, was
chosen as a cable. The maximum cable length is 25 meters (half- or full-duplex). Two connectors are
specified for 1000BASE-CX. Because the 1000BASE-X physical layers were derived from the Fibre
Channel standard, it made sense to use the same connector: the High-Speed Serial Data Connector
(HSSDC), commonly known as Fibre Channel Style 2 connector. The IEEE also specified the nine-
pin D-subminiature connector, which is also used for Token Ring and the STP version of 100BASE-
TX.

1000BASE-T: Gigabit Ethernet for Four-Pair Category 5 UTP

The IEEE 802.3ab committee is completing a new standard that defines Gigabit Ethernet
transmission over four-pair Category 5 or better cable. 1000BASE-T, as the standard is known, is an
incredible engineering feat. When 100MHz Category 5 cabling was first conceived, the maximum
data rate was considered to be around 100Mbps. The IEEE engineers have managed to squeeze 10
times that out of Category 5 cabling. It is worth spending a few minutes on how they propose doing
this. 1000BASE-T combines several tricks to achieve 1000Mbps over Category 5:

l 1000BASE-T uses all four pairs of Category 5 cabling. You can still use a regular Category 5
RJ-45 connector.

l 1000BASE-T uses simultaneous transmission and reception on all four pairs. All other physical
layers in use today either transmit or receive, but never both. This simultaneous transmission

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 23 of 36

from both ends is sometimes called dual-duplex.

l To make better use of the available bandwidth, 1000BASE-T uses the same encoding scheme
that was developed for 100BASE-T2, namely PAM5. 1000BASE-T operates at 125MHz, the
same frequency as 100BASE-TX, which is five times that of 100BASE-T2. Because
1000BASE-T is essentially a combination of 100BASE-TX and 100BASE-T2, this encoding
scheme is also called enhanced TX/T2.

l 1000BASE-T combines many existing technological features, but the combination of four
simultaneous transmission and reception on all four pairs, the complex PAM5 encoding, and
the high symbol rate of 125MHz make it a state-of-the-art transmission technology. This
requires a highly sophisticated DSP to make it actually work. The PHY transceiver chip at both
ends of the cable will actually be a complex DSP engine that will contain more than 200,000
transistors, about the same as an Intel 486 microprocessor!

NOTE
DSP is an abbreviation for digital signal processor, a specialized microprocessor
designed to analyze and process analog signals.

The 1000BASE-T standard should be completed in 1999, at which point products are also expected
to start shipping.

Table 3.10 summarizes the four different Gigabit Ethernet physical layers.

TABLE 3.10 A SUMMARY OF THE DIFFERENT GIGABIT ETHERNET PHYSICAL

LAYER IMPLEMENTATIONS

PHY Parameter 1000BASE-T 1000BASE-SX 1000BASE-LX 1000BASE-

CX
IEEE Standard 802.3ab1 802.3z (July 1998)
Number of pairs Four pairs Two strands Two strands Two pairs
required
Cable Category Category 5 or 50 or 62.5 um 50 or 62.5 um MMF or 8- 150-W
required better MMF 10mm SMF Twinax
Cable length 100m 220-550m 5000m (SMF) 550m 25m
(MMF)2
Encoding 4D-PAM5 8B/10B 8B/10B 8B/10B
Connector specified RJ-45 SC SC HSSC or DB-
9
Proven Not yet 3 3 3
Full-duplex capable Yes 3 3 3
Broad product None shipping 33 3 Probably not3
support yet1

1 802.3ab approval is expected in mid-1999. Products are expected around the same timeframe.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 24 of 36

2 Requires special DMD patch cord. See Chapter 6 for more details.

3 By September 1998, not a single Gigabit Ethernet vendor was shipping CX-based product.

The IEEE Auto-Negotiation Standard

When 100BASE-T was invented, it was no longer safe to assume that a typical Ethernet RJ-45
connector would be carrying 10BASE-T. Instead, a UTP cable could carry one of seven different
Ethernet signals: 10BASE-T, 10BASE-T full-duplex, 100BASE-TX, 100BASE-TX full-duplex,
100BASE-T4, 100BASE-T2, or 100BASE-T2 full-duplex.The IEEE incorporated a very elegant
scheme into the 802.3u standard that has simplified your life as a LAN manager tremendously. The
IEEE’s auto- negotiation technology, originally developed by National Semiconductor and known as
NWay, can tell what speed the other end of the wire can handle. The repeater, switch, or NIC will
then automatically adjust its speed to the highest common denominator--the fastest speed that both
can handle. This is how auto-negotiation works:

l Both devices need to contain the auto-negotiation logic. (All Fast Ethernet equipment shipping
after mid-1996 includes this feature.)

l Auto-negotiation is an enhancement of the 10BASE-T link integrity signaling method and is

backward compatible with link integrity.

l New equipment incorporating the auto-negotiation feature still allows you to select one of the
possible modes manually. NICs typically do this via a command-line option; switches can be
set via the management software.

This technology has several benefits. Assume that you are working in someone’s office, connecting
his or her PC. You can tell that the PC has an older ISA 10BASE-T NIC installed. You want to
connect the PC to the RJ-45 Ethernet outlet via a UTP patch cord, but you are not sure what hub the
outlet is connected to in the wiring closet. It could be a newer 10/100BASE-TX hub, or an older
10BASE-T hub. Without auto-negotiation, you would have to walk back to the wiring closet to find
out because connecting a 10BASE-T NIC to a 100BASE-TX hub could cause some serious problems,
because the electrical signal levels are incompatible. Auto-negotiation will eliminate this possibility;
it will not allow dissimilar technologies to connect or interfere with each other. If this outlet has a
10BASE-T hub connected to it, everything is okay. If you do indeed have a 10/100TX hub installed,
it will auto-negotiate its speed down to 10Mbps.

Another scenario would be when you wanted to upgrade a user from a 10Mbps connection to
100Mbps. Let’s say you had the foresight a few years ago to install 10/100TX NICs everywhere. Now
you have bought some new 10/100 workgroup switches. All you need to do is swap out the old
10BASE-T hub for the new 10/100 switch, and the upgrade is done. All the NICs will sense that they
are now connected to a 100Mbps switch; you do not need to visit each workstation.

Auto-negotiation uses a series of fast link pulses (FLPs), similar to the 10BASE-T link integrity (LI)
pulses. Both hub and NIC send out this sequence of pulses, which allows the other end of the wire to
identify the type of Ethernet connection the host is capable of (see Figure 3.12).

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 25 of 36

Because different Ethernet devices can support multiple Ethernet functions, a prioritization scheme
exists to make sure the highest or fastest available option is chosen. This scheme will ensure that a
10/100 full-duplex device will always prioritize 100Mbps over 10Mbps operation, and full-duplex
over half-duplex, and so on. Table 3.11 illustrates this prioritization.

100BASE-T2 is prioritized ahead of 100BASE-TX and 100BASE-T4 because 100BASE-T2 runs

across a broader spectrum of copper cabling and can support a wider base of configurations.
100BASE-T4 is ahead of 100BASE-TX because 100BASE-T4 runs across a broader spectrum of
copper cabling. Full-duplex is always ahead of half-duplex because if the capability exists you want
to take advantage of it. 10BASE-T is last because it runs at the slowest speed.

FIGURE 3.12 The auto-negotiation FLP is similar to the 10BASE-T link integrity pulse and encodes
information to determine the highest possible speed.

TABLE 3.11 AUTO-NEGOTIATION DEFINES A PRIORITY HIERARCHY FOR

ENSURING THAT TWO CONNECTIONS COMMUNICATE AT THE HIGHEST
COMMON SPEED

Priority Comment
1 1000BASE-T full-duplex1

2 100BASE-T2 full-duplex
3 100BASE-T2
4 100BASE-TX full-duplex
5 100BASE-T4
6 100BASE-TX
7 10BASE-T full-duplex
8 10BASE-T

1 1000BASE-T also uses the same RJ-45 connector. We expect the auto-negotiation scheme to be
modified to include 1000BASE-T full-duplex as the number one priority.

Fiber-optic auto-negotiation also exists. Most fiber-optic devices only support one mode, however,
and it is not possible at this point to build dual-speed devices. The only benefit of auto-negotiation for
fiber devices is to arbitrate between half- and full-duplex.

Before the IEEE officially adopted auto-negotiation as the part of the 100BASE-T/802.3u standard,
some vendors were shipping proprietary auto-sensing network adapters. A lot of 10/100 NICs
installed out there are running at 10Mbps today, to be upgraded to 100Mbps at some point in the
future. Table 3.12 illustrates how different hubs and NICs will interoperate with and without the auto-
negotiation scheme.

TABLE 3.12 INTEROPERABILITY OF PRESTANDARD AUTO-SPEED AND AUTO-

NEGOTIATION SWITCHES AND NICS

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 26 of 36

1st -Generation New Auto-

10/100 Hub (No Auto- Negotiation
NIC-Only Hub 10BASE-T- 100BASE-TX- Sensing)1
Only Hub 10/100TX Hub
10BASE-T-only No choices canN/A (better buy Hub manually set to 10 Hub auto-
NIC be made. a new NIC!) mode (with negotiates to 10
management software). mode.
Pre-standard auto- NIC NIC Manually set hub to Manually set hub
sensing10/100TX automatically automatically 100 mode, NIC will and NIC to 100
NIC 2 selects 10 mode. selects 100 automatically adjust to mode.
mode. 100 mode also.
New auto- NIC auto- NIC auto- Manually set hub and Both hub and
negotiation10/100TX negotiates to 10 negotiates to100 NIC to 100 mode. NIC auto-
NIC3 mode. mode. negotiate to 100
mode.

1 Forexample, the SynOptics 28115 hub will operate at either 10 or 100Mbps, but does not
automatically adjust its speed. The speed adjustment has to happen manually via the switch
management software.

2 Examples are the Intel EtherExpress Pro/100, or the 3Com Fast EtherLink.

3 The Intel EtherExpress Pro/100 Model B and the 3Com EtherLink XL include auto-negotiation.

Ethernet Repeaters
Repeaters extend the size of a network by joining multiple segments into a larger segment. A repeater
works at the physical layer (Layer 1) of the OSI model. That means the repeater does not process the
data at all. A repeater has no MAC, only PHYs, and merely receives the incoming signals and
reconditions them for immediate retransmission on all ports. Repeaters are synonymous with shared
media. A repeater is invisible to all nodes on a repeated LAN. It appears as though all nodes are
connected via one cable.

Repeaters are almost as old as Ethernet itself. Fast and Gigabit Ethernet make provisions for
repeaters, although the trend toward dedicated media and switching has been accelerating for the past
five years. With Gigabit Ethernet in particular, repeaters might soon be outdated. Figure 3.13 shows
the internal workings of a repeater.

FIGURE 3.13 A repeater operates at Layer 1 of the OSI model: it only contains PHYs and some
core logic, but no MACs. This figure shows a repeater receiving a degraded signal on port 1 and
retransmitting a clean signal on ports 2-4.

How Repeaters Work

All Ethernet repeaters work as follows:

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 27 of 36

1. An encoded data signal is received on a particular port, either from a node, a switch, or
another repeater. This signal has been electrically degraded because it has traveled some
distance from the source to the repeater over less-than-perfect cable. (This applies to both UTP
and fiber; the degradation just takes longer over fiber.)

2. The incoming port, PHY, processes the data and re-creates the perfect digital data internally.

3. Finally, the digital signal is forwarded to all ports, where the PHY on each port converts the
data into an encoded signal, to be transmitted over the wire again.

Note that no data is stored, no frames are re-created, and every received bit is sent on, irrespective.
Figure 3.14 illustrates the flow diagram of a repeater.

FIGURE 3.14 Flow diagram of the repeater forwarding logic: There isn’t much to it. Later on you
will see the same diagram for a switch, which is much more complex.

The Repeater Collision Domain

A segment is defined as a group of nodes connected to the same repeater. Multiple segments can be
connected via repeaters. A repeater propagates all network traffic present on one segment to all other
segments to which it is connected. All interconnected segments are in one electrical collision domain.

Electrical signals take a certain time to travel across a cable. In addition, all repeater hops introduce a
small delay or latency. This is the delay between the time an incoming signal is received and the time
that signal is transmitted again to all ports. These two forwarding delays and their impact on Ethernet
collision detection is a key factor for determining 10, 100, and 1000Mbps Ethernet network design
and diameter rules.

A collision occurs when two nodes simultaneously sense that the media is available and then attempt
to transmit at the same time. Somewhere in the shared collision domain the two data patterns literally
collide, causing a voltage surge on the cable. The voltage surge travels back along the cable to the
two sources, where it is detected by the transmitters. A collision must be detected by the nodes
causing the collision before they stop transmitting, including the one at the farthest end of the cable;
otherwise, the nodes would never know that their transmission had been corrupted and would proceed
as though nothing had happened. After the transmitting station recognizes that a collision has
occurred, a JAM signal is generated, which is equivalent to canceling the last transmission. The worst
case for collision detection is very short frames, because they leave very little time for the collision
signal to travel back to the transmitter. Figure 3.15 shows the collision domain and minimum frame
size.

Let’s look at how the maximum network diameter depends on the minimum frame size. First, we
need to define a variable called the transmission time. This is the amount of time that the
transmission takes, from beginning to end. It is also equal to the time that any collision has to travel
back to the transmitting station.

The minimum transmission time can be calculated from the minimum frame size. For 10Mbps
Ethernet, the minimum transmission time is

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 28 of 36

The signal needs to travel from the transmitter to the point where the collision occurs. Then the
collision signal needs to travel back to the transmitter again. The minimum transmission time number
in the preceding equation needs to be divided by 2 to account for round-trip delays. Therefore, the
minimum transmission time is equal to 25.6ms (round-trip).

FIGURE 3.15 Collisions occur when two nodes attempt to transmit simultaneously. The maximum
network diameter is directly related to the minimum frame size.

Copper and fiber cable have a delay of between 0.5 and 0.6ms/100m. To calculate the maximum
network diameter, we merely need to divide the minimum transmission time of 25.6 ms by the cable
delay.

This calculation still excludes any delays introduced by repeaters, nodes, and so on. Many
manufacturers specify the latency of their repeaters, so you could calculate the exact network
diameter depending on the manufacturer’s numbers. This is very cumbersome. To make our lives a
little easier, the IEEE came up with a quick and easy-to-remember rule that limits the 10Mbps
network size to 2500 meters. The rule is described in more detail later.

When 10Mbps Ethernet was originally designed, the minimum frame size of 512 bits was chosen to
come up with a realistic network diameter. The problem with Fast Ethernet and Gigabit Ethernet is
the bit time. The network diameter is directly related to the minimum frame size, as well as the bit
time.

As we move from 10 to 100 to 1000Mbps, the bit time decreases by an order of 10 every time. If we
have a network diameter of 2500 meters for 10BASE-T, this shrinks to one-tenth, or approximately
250 meters for 100BASE-T, which is already marginal. For Gigabit Ethernet, the network diameter
would shrink to 25 meters, which is clearly not a viable option. Therefore, the IEEE decided to
increase the minimum frame size for Gigabit Ethernet to 512 bytes (or 4096bits), as opposed to 64
bytes for 10 and 100Mbps Ethernet. This, as you might notice, is only an eight-fold improvement.
This yields a network diameter of around 200 meters for Gigabit Ethernet. Table 3.13 summarizes the
different collision diameter variables for 10, 100, and 1000Mbps Ethernets.

You can calculate the exact maximum network diameter for any network using the delays for cable
and manufacturers specifications for latencies, but nobody would really want to do that. (For you die-
hards who insist, we have put down some guidelines. See the section "Calculating Your Own
Collision Diameter," later in this chapter.)

The IEEE made things easy and developed a set of guidelines. We discuss these cheat sheet
guidelines for 10, 100, and 1000Mbps repeater segments in the following section.

10Mbps Ethernet Repeater Rules

You can easily memorize the 10Mbps Ethernet golden rule as the 5-4-3-2-1 rule:

l Five segments are allowed (of 500-meter diameter each).

l This implies four repeater hops in the data path.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 29 of 36

l Three of these segments may be populated with nodes.

l Two segments cannot be populated but are only inter-repeater links.

l All of this makes one large collision domain with a maximum of 1024 stations. Total network
diameter can be up to 2500 meters.

TABLE 3.13 THE COLLISION DIAMETER FOR 10, 100, AND 1000MBPS ETHERNETS
ARE A FUNCTION OF MINIMUM FRAME SIZE AND BIT TIME

Parameter 10Mbps 100Mbps 1000Mbps

Minimum frame size 512 bits(64 bytes) 512 bits(64 4096 bits (512
bytes) bytes)1
Collision diameter (bit times)2 512 bit times 512 bit times 4096 bit times
Bit time (ms)3 0.1ms 0.01ms 0.001ms
Maximum round-trip delay (ms)4 51.2ms 5.12ms 4.096ms
Maximum network diameter, without Approximately45710m 457m(1/10 of 3661 m (8/10 of
repeaters 5 Ethernet) Fast Ethernet)
Maximum IEEE-specified network 2500m 205m 200m
diameter, with 100-meter UTP
connections 6
Maximum number of repeaters with ~5 2 or 1 1
specified network diameter7

1 This number is eight times the regular or Fast Ethernet minimum frame size. This doesn’t entirely
make up for the one-tenth reduction in diameter due to the speed increase, but it comes close. To be
totally correct, the actual Ethernet frame size has not increased. Instead, the minimum Gigabit
Ethernet frame size is still 64 bytes, just like 10 and 100Mbps Ethernet frames. The 802.3z
specification requires a carrier extension to be added to small frames to keep the wire busy for a
minimum of 512 bytes. In this way, it looks like the minimum frame size is 512 bytes.

2 Collision diameter = minimum frame size. The IEEE specifies the minimum frame size.

3 Bit time = 1/data rate. Depends on the speed of Ethernet.

4 Round-trip delay = bit time * minimum frame size.

5 Absolute best case, calculated using a copper delay of 0.56ms/100m segment. Ignores 100m UTP
limitation. Also assumes no repeater delays, so this is a theoretical number. Because fiber cabling
has similar delays to copper, this number comes close to the half-duplex fiber network diameter
specifications.

6 This is the real-world number that includes a budget for repeaters, too.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 30 of 36

7 We discuss exact repeater rules later in the chapter.

Figure 3.16 illustrates this 5-4-3-2-1 design rule, which applies to 10Mbps Ethernet only. This 5-4-3-
2-1 rule is only a rough guideline, but it works very well in most cases. Actual numbers vary by
manufacturer. We have seen from four to seven repeater hops being specified. (In theory, as discussed
previously, a 10Mbps Ethernet LAN could have a collision diameter of at least 4000 meters, if no
repeaters were present. This is a number that can be attained only by using a point-to-point single
fiber connection because all copper-based cabling will need to be repeated multiple times to go this
far.)

FIGURE 3.16 The Ethernet 5-4-3-2-1 repeater rule stipulates a maximum of five repeaters, four
segments, three populated, and two repeater links, making up one collision domain.

100BASE-T Repeater Rules

The IEEE repeater rules have hence been significantly changed for 100BASE-T. The IEEE 802.3u
Fast Ethernet standard contains some specifications for Fast Ethernet repeaters. Due to the faster
speed, the Fast Ethernet network diameter has been reduced from 2500 meters to approximately one-
tenth that size.

The actual diameter depends on two things:

l The type of cable used--Fiber cable has a slightly smaller delay: about 0.5ms/ 100, as opposed
to UTP, which has a delay of 0.56ms/100m. (Note that all UTP segments still need to be 100
meters, as this is an EIA/TIA spec.)

l The type of repeater used--With 10BASE-T, latency was not an issue, but with 100BASE-T, it
becomes a major differentiating feature for a repeater. As the network diameter is rather limited
for Fast Ethernet, the IEEE has defined two different classes of repeaters, called Class I and
Class II:

l A Class I repeater can have a relatively large port-port timing delay of 0.7ms or less. It operates
by translating line signals on an incoming port to digital form and then retranslating them to
line signals when sending them out on the other ports. This makes it possible to repeat signals
between different media segments, such as FX and TX, within the same repeater hub. The
larger delay means the 2.56ms collision diameter only allows for one Class I repeater to be
used with two 100-meter cable lengths. For two fiber links, the maximum cable diameter is 272
meters.

l Class II repeaters have a lower latency of 0.46ms or less. Class II repeaters do not regenerate
the digital signal like a Class I repeater does. A Class II repeater immediately repeats the
incoming signal to all other ports without a complete translation process. Class II repeaters are
preferred from a network design perspective because the lower latency allows for two repeater
hops. The drawback of Class II repeaters is a single repeater cannot mix different media PHYs.

l Class I and Class II differ with respect to their internal design and latency characteristics. This
allows a more accurate network diameter calculation to get closer to the limit.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 31 of 36

If two 100-meter UTP links are used, this configuration allows for an inter-repeater link of 5-meter
UTP to yield a total network diameter of 205 meters. Alternatively, two fiber connections and an
inter-repeater fiber link with a total length of 228 meters are permitted. Most Fast Ethernet repeaters
shipping today are Class II.

The two cases just described both use only one media type for the entire network diameter. The IEEE
has also laid down some guidelines for mixed copper-fiber installations, but these situations are rare.
Table 3.14 illustrates the network diameter guidelines for these situations. Figure 3.17 graphically
depicts Table 3.14.

TABLE 3.14 FAST ETHERNET COLLISION DIAMETER RULES

Connection All UTP Mixed-Media 100BASE-TX1 All

and 100BASE-FX Fiber
Repeated Network
Segment Diameter
One Class II repeater 200m 309m 320m
One Class I repeater 200m 261m 272m
Two Class II repeaters 205m 216m 228m
Point-to-Point
Connections2
Node-node or switch- 100m N/A 412m
switch half-duplex
Node-node or switch- 100m N/A 2000m
switch full-duplex

1 T4 allows for a few meters more.

2 Wehave shown the point-to-point distance limitations for reference only. For single-mode fiber
connections operating in full-duplex, even longer segments are possible, but the IEEE didn’t specify
SMF as part of the Fast Ethernet standard.

FIGURE 3.17 The network diameter for a 100BASE-T network can range from 200 meters for UTP
to 320 meters for fiber, assuming one repeater.

Some people originally viewed the Fast Ethernet repeater specification (two repeater hops and a 205-
meter UTP network diameter) as a big issue. Most people soon realized, however, that one could
effectively work around both issues with the use of stackable repeaters and switching hubs to get to
any network diameter and an infinite number of hops. Refer to the deployment chapters for more
details.

Five years ago, a network diameter of 205 meters combined with one repeater hop would have made
100BASE-T impractical if not altogether useless for most LAN managers. Two technological
developments, however, have made it possible for 100BASE-T repeaters to work well even within
these tight restrictions:

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 32 of 36

l Most 100BASE-T networks are additions to an existing 10BASE-T network. Every new
100BASE-T segment that is added to a 10Mbps network will require a switch to get from
10Mbps to 100Mbps. The 205-meter calculation is started at the switch, and 205 meters is
sufficient to get to most nodes with one or two intermediate repeater hops. If the distance from
switch to node is longer than 205 meters, another switch must be added to further extend the
network diameter. Alternatively, 100BASE-FX could be run to repeaters or nodes that are
longer than 205 meters away from the switch.

l Standalone, unmanaged 100BASE-T repeaters will be rare. Most 100BASE-T repeaters will be
stackable, meaning that many repeaters can be physically placed on top of each other and
connected via a fast backplane bus. The fast backplane bus does not count as a repeater hop and
makes the entire stack look like one larger repeater. In fact, most 100BASE-T stackables can be
stacked four or more high. Electrically, the repeater stack appears as one larger repeater.

Calculating Your Own Collision or Network Diameter

You can calculate your own network diameter if you like. Use the following equation to make sure
your repeated Ethernet network segment will function satisfactorily:

Note the following:

l You can obtain the maximum round-trip delay from Table 3.15. It is 51.6ms for 10Mbps
Ethernet, 5.12ms for Fast Ethernet, and 4.096ms for Gigabit Ethernet.

l The factor 2 accounts for round-trip delays as the frame needs to travel from transmitter, to the
collision point and back to the transmitter again.

l We can obtain repeater and NIC delays from the manufacturer. Ethernet repeaters typically
have latencies of 2ms or less. Fast Ethernet Class I repeater delays are less than 0.7ms; Class II
delays are less than 0.46ms. A 10Mbps NIC delay is about 1ms; Fast Ethernet NIC delay is
0.25ms. All networks are terminated on both ends: two NICs or a NIC and a switch port.

l We can obtain the cable delays from Table 3.15. Cable delays do not at all depend on the
Ethernet speed. Cable delays typically are measured as a fraction of the speed of light and
mostly depend on the insulation material used. A 100-meter UTP section still takes about
0.55ms one way.

l For 10Mbps networks, 51.2ms allows for quite a few repeaters, cable segments, and DTEs to
exist without exceeding the collision domain restrictions. In most real-world 10BASE-T
networks, you will find it very difficult to exceed this number.

TABLE 3.15 INDIVIDUAL NETWORK COMPONENT DELAYS FOR CALCULATION OF

NETWORK DIAMETER

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 33 of 36

Component Maximum Delay Times in ms

Two Fast Ethernet NICs or switch 0.5
ports
Fast Ethernet MII 0.2
100m Category 5 cable segment 0.556
1m Category 5 cable 0.00556
100m Fiber optic cable segment 0.5
1m fiber cable 0.005
Fast Ethernet Class I repeater 0.7 max
Fast Ethernet Class II repeater 0.46 max
Two Gigabit Ethernet NICs 0.864 max
Gigabit Ethernet repeater 0.488 max

Figure 3.18 illustrates the different delay components in a shared-media network.

FIGURE 3.18 If you want to calculate your own maximum network diameter, you need to add up the
delays of the individual components and make sure they don’t exceed half the minimum frame size.

Some additional points to note:

l If you are using an MII cable, you need to add it to the cable segment length.

l Decide on an appropriate safety margin. (We recommend five bit times.)

l You will see that it is actually possible to build 100BASE-T networks with three or even four
repeater hops if the individual cable segments are significantly less than 100 meters.

Gigabit Ethernet Repeater Rules

As mentioned previously, with CSMA/CD, the capability of a station to detect collisions directly
relates to the round-trip propagation delay on the cable. This variable is in turn related to the original
64-byte minimum frame size for Ethernet, which yielded a practical network diameter of 2500
meters. With Fast Ethernet, the 64-byte minimum frame size was kept the same, but the network
diameter was reduced from 2500 meters to 205 meters (Category 5 UTP).

Some people in the IEEE wanted to drop half-duplex and, therefore, CSMA/CD altogether from the
Gigabit Ethernet specification. This didn’t fly, and the IEEE decided that Gigabit Ethernet should still
have a half-duplex repeater option. Therefore, the IEEE had two choices. First, the IEEE could keep
the frame size the same and reduce the network diameter again by a factor of 10, which would have
meant a network diameter of merely 20 meters. Alternatively, the IEEE could increase the frame size
to increase the network diameter again. The IEEE opted for the latter and increased the minimum
frame size to 512 bytes. This minimum frame size allows for a practical network diameter of between
200 and 220 meters, depending on the cable used. Table 3.16 summarizes the IEEE Gigabit Ethernet
network diameter rules.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 34 of 36

TABLE 3.16 GIGABIT ETHERNET NETWORK DIAMETER SPECIFICATIONS

Media Maximum Collision Diameter

Point-to-Point, Half-Duplex
1000BASE-CX node-node or switch- 25m
switch1
1000BASE-T node-node or switch- 100m
switch1
1000BASE-SX or LX node-node or 316m switch-switch1
One Repeater Segment
1000BASE-CX 50m
1000BASE-T2 200m
1000BASE-SX or LX 220m

1 Remember these distances are half-duplex. For full-duplex operation, see Table 3.9.

2 Mixed-media repeater limitations are 210 meters for UTP/fiber, 220 meters for twinax/fiber.

The Gigabit Ethernet repeater rule is simple: 1 repeater, 2 UTP, or fiber segments of 100 meters each.
It is very unlikely, however, that you will ever need to apply this 802.3z IEEE Gigabit Ethernet rule.
That’s because you probably won’t ever see a true Gigabit Ethernet repeater for sale. At the time of
this writing, no Gigabit Ethernet repeaters exist. Maybe the 1000BASE-T standard will change that,
but we doubt it because the new 802.3X full-duplex flow/control standard, approved in March 1997,
has effectively made half-duplex Ethernet redundant. Instead of CSMA/CD repeaters, vendors are
building a hybrid device called a full-duplex repeater. This device looks like a repeater, but doesn’t
have some of the disadvantages of regular repeaters.

Gigabit Ethernet Full-Duplex Repeaters

Gigabit Ethernet vendors are building a new class of device, called the full-duplex repeater. This
device is also known as a buffered distributor, nonfiltering switch, or CSMA/CD in a box. We like to
use the term full-duplex repeater because the device acts like a repeater.

The full-duplex repeater combines the cost-effectiveness of a repeater with the performance
characteristics of a switch. The device is a full-duplex, multiport, repeater-like device. Regular
repeaters operate only in half-duplex mode; full-duplex repeaters and buffered distributors, on the
other hand, feature full-duplex links on all ports and a MAC and a PHY on every port. We discuss
these devices in more detail in the next chapter, after we examine switches in general and the 802.3x
flow-control/full-duplex standard more closely.

Here are the key characteristics of a full-duplex repeater:

l Like a repeater, the full-duplex repeater forwards all incoming packets to all connected links
except the originating port. That’s the shared bandwidth repeater part.

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 35 of 36

l All the ports operate at the same speed, 1Gbps in this case. The full-duplex repeater concept,
however, is not at all linked to wire speed.

l The full-duplex repeater has a very rudimentary MAC function on every port. The MAC will
make sure a complete frame is received, perform the FCS to make sure the frame is valid, and
then buffer it in internal memory. Full-duplex repeaters have enough memory capacity for a
small number of frames to be buffered.

l The full-duplex repeater also looks a little bit like a switch. All connections use full-duplex
links, so the full-duplex repeater can simultaneously transmit and receive on an individual port.

l The full-duplex repeater uses flow control to make sure that the internal buffers don’t overflow.
When memory begins to fill, the full-duplex repeater invokes 802.3x flow control to inform the
transmitting node to stop sending while it empties its buffers.

The following are some important points that differentiate a full-duplex repeater from a switch:

l Switches forward only to the appropriate port, not to all ports like the full-duplex repeater.

l Switches typically have much larger internal memory capacity. Full-duplex repeaters have
storage for just a few frames at most.

l Switches have backplanes that support bandwidths greater than the individual port rate to
accommodate multiple data streams simultaneously. The backplane of a full-duplex repeater is
exactly equal to the data rate.

l A full-duplex repeater will not work properly unless all transmitting nodes comply with 802.3x
and operate at the same wire speed, which is Gigabit Ethernet in this case. A switch, on the
other hand, can typically accommodate different data rates and half- or full-duplex nodes.

l A full-duplex repeater does not examine the contents of the frame beyond the FCS; no source
or destination address analysis is done. The device blindly forwards all good frames to all ports
(except the originating one).

Figure 3.19 shows an internal block diagram of a full-duplex repeater.

FIGURE 3.19 The new Gigabit Ethernet full-duplex repeater is sending and receiving via its buffer
memory on all ports. Internally, the round-robin logic is busy forwarding the data stored in port 1’s
receive buffer to the transmit buffers for ports 2-4.

Summary
This chapter covered Layer 1 and Layer 2 aspects of the three different versions of Ethernet. The
Ethernet MAC operates at the data link layer (Layer 2) of the OSI model. The Ethernet MAC utilizes
a transmission algorithm called CSMA/CD.

The Ethernet MAC has scaled by a factor of 100 over the past 20 years, originally starting at 10Mbps,

file://I:\chapters\z\zd594.html 3/21/01
Switched, Fast, and Gigabit Ethernet - CH 3 - Ethernet, Fast Ethernet, and Gigabit Ethern.. Page 36 of 36

then speeding up to 100Mbps, and now 1000Mbps. A 10Gbps version is technically very feasible.

We looked at the different physical layers (OSI Layer 1) associated with each different Ethernet speed
version. 10Mbps Ethernet typically utilizes UTP, known as 10BASE-T. 10Mbps Ethernet started off
as a coax-based networking technology, known as 10BASE5 or 10BASE2. These days, 10Mbps
Ethernet typically runs on UTP or fiber (10BASE-T or 10BASE-F).

The 100Mbps Ethernet standard encompasses four different physical layers, with the prevalent ones
being 100BASE-TX and 100BASE-FX. 100Mbps Ethernet is also known as Fast Ethernet.

Gigabit Ethernet, the latest incarnation of Ethernet, can be run over four different cabling types. The
1000BASE-SX and LX fiber variants cater to many different qualities of fiber. The future of the
shielded twisted-pair version, known as 1000BASE-CX, is questionable. The new 1000BASE-T
standard will utilize existing Category 5 cabling for 1000Mbps transmission.

Ethernet started off as a shared-media technology. Repeaters are used in a shared media network to
increase the network size. As the speed of Ethernet has increased, the network diameter has had to
shrink proportionally. With 10Mbps Ethernet, at least five repeater hops are possible. With 100Mbps
Fast Ethernet, this had to be reduced to one or two. Gigabit Ethernet employs a trick called carrier
extension to even make one repeater hop possible.

We most likely will never see any Gigabit Ethernet shared-media hardware. Over time, technological
enhancements have enabled Ethernet to become a full-duplex, switched technology. In essence, this
means Ethernet is no longer using the CSMA/CD access method. Only the Ethernet framing format
remains the same.

file://I:\chapters\z\zd594.html 3/21/01