Tải bản đầy đủ
7 Alternative physical layers—ethernet, fast ethernet and gigabit ethernet
Alternative physical layers — ethernet, fast ethernet and gigabit ethernet
The physical layer signalling (PLS) sublayer controls the carrier sensing and reacts to
collision detection as we described earlier. The AUI (attachment unit interface) passes signals
from the PLS to the PMA (physical medium attachment). In practice, AUI is a cable with
DB-15 plugs and sockets. Such connectors (labelled AUI) are still found on older networking
equipment and ethernet network interface cards (NIC). AUI also defines the coding of the
physical layer signal to be Manchester coding (see Chapter 2–Figure 2.18).
The physical medium attachment (PMA) is achieved using a device known as a medium
attachment unit (MAU). It is the PMA which is responsible for the actual detection of collisions,
notifying them by means of the AUI to the PLS for action. It also regulates when transmissions
may be sent onto the medium, but otherwise merely forwards the already line-coded signal,
adapting it for the actual type of coaxial cable or other medium in use. In the later days of
coaxial cable networks, the MAU was a connection device which could be incorporated into the
cable network itself by means of a BNC socket. Alternatively it was sometimes incorporated
into wall sockets (behind which was the coaxial cabling of the bus). The medium dependent
interface (MDI) in this case is the specification of one of the different allowed 50 coaxial
cable types (thicknet, thinnet) and the associated BNC (bayonet connector) connectors.
In modern 10baseT ethernet, it is normal for the AUI and MAU functionality to be combined
into the network interface card (NIC). This is both cheaper and reduces the possible sources
of failures. The standard interface format from the NIC is nowadays an RJ-45 socket. An
RJ-45 category 5 patch cable is used to connect the NIC to a similar socket on the LAN hub.
Should, however, a modern NIC be required to be connected to an older coaxial cable ethernet
or standard AUI (DB-15 connector) then a transceiver is used to do the conversion. This is a
small device with a single RJ-45 socket on one side, and an AUI interface (DB-15 connector or
BNC connector) on the other — for direct connection to the coaxial cable ethernet backbone.
Fast ethernet physical layer (100baseT, IEEE 802.3u)
The active hubs (switches) used in fast ethernet networks make it possible to combine the
different types of fast ethernet and older 10baseT ethernet devices in a single network. Such
backward-compatibility was given high importance by fast ethernet designers. Naturally therefore, the combined 10/100baseT hubs have to cope with much more than the passive hubs of
simple 10baseT networks. First of all, each of the ports of a 10/100baseT active hub may be
running at different speeds (either 10 Mbit/s or 100 Mbit/s). Second, each port speed may first
have to be either autosensed (the hub adjusts to the speed of the device) or auto-negotiated
between the devices (which discover which particular fast ethernet technology is in use (TX,
There were new considerations to be taken care of in the specifications of fast ethernet:
• The higher bit rate demanded faster interfaces between the protocol layers and on the
• Backward compatibility was essential, in order that older DTE devices with existing ethernet cards could coexist with newer fast ethernet devices in the same LAN.
• Full duplex operation was defined, since point-to-point connections of fast ethernet were
envisaged as backbone links between different ethernets.
At the time of introduction of fast ethernet, modifications had to be made to the ethernet MAC
and physical layers to take account of the much higher speed of transmission. Fast ethernet
uses technology for high speed physical data transfer which came from FDDI (fibre distributed
Local area networks (LANs)
data interface).6 The physical layer technology of FDDI was simply adapted to interface the
existing ethernet PLS/MAC interface. At the same time, the old sublayers (reflecting the
coaxial cable heritage of basic ethernet) were dispensed with, and a new sublayer model was
born (see Figure 4.9).
Backward compatibility with the 10baseT version of ethernet demanded a similar MACframe structure (Figure 4.5 and Table 4.4) to that of 10 Mbit/s ethernet and a similar serial
interface (1-bit at a time) from the MAC layer. This was achieved with a new reconciliation
The MII (medium independent interface) of fast ethernet replaces the AUI of 10 Mbit/s. The
MII is the interface which carries the physical layer signalling (PLS) to the physical medium
itself and provides a standard medium-independent interface for this interface. It comprises
4 separate transmit and receive wires, each operating at 25 Mbit/s (for fast ethernet) or at
2.5 Mbit/s (for 10 Mbit/s ethernet). This enables four bits at a time (a nibble) to be carried
by the MII at a lower baud rate. As discussed in chapter 2, such multilevel transmission
allows a higher bit rate at a lower baud rate by sending multiple bits at once using a single
symbol. Instead of using Manchester coding, the MII uses NRZ (non return-to-zero) coding.
The reconciliation sublayer (RS) converts to the ‘old’ one-bit-at-a-time interface required by
the MAC. Both MII and RS also are adapted to support full duplex operation. If present as a
cable interface (like AUI), the MII cable may be a maximum of 0.5 m long and has 18 pins
(4 transmit, 4 receive, 10 control).
A new layer, the physical coding sublayer (PCS) performs similar functions to that of the
10 Mbit/s physical layer signalling (PLS) sublayer. Meanwhile, the physical medium attachment (PMA) sublayer converts the standard internal format of MII (a parallel format like that
shown in Figure 2.26) into one of the different physical layer alternatives, the connectors, pin
layouts, line codes and physical properties of which are defined by the remaining layers: PMD
(physical medium dependent) and MDL (medium dependent layer).
There are several alternative physical forms of fast ethernet, the most important of which are:
• 100baseTX, which uses 2-pair category 5 cabling like 10baseT, MLT-3 (multi-level transmission 3) line coding (a multiple state higher level transmission technique requiring only
32 MHz bandwidth: category 5 unshielded twisted pair [UTP] cable can only provide
100 MHz of bandwidth);
• 100baseFX, which uses a single fibre pair, together with a combination of 4B/5B block
code and non-return-to-zero-inverted (NRZI) line code;
• 100baseT4, which uses 4-pair (2 pairs for transmit and 2 pairs for receive) category 3, 4
or 5 cabling and 8B/6T line coding (This technique overcomes the limited bandwidth of
the UTP cable by using multiple cable pairs for the transmission); and
• 100baseT2, which is a half-duplex version of 100baseT similar to 100baseT2 but requiring
only 2 pairs.
An optional auto-negotiation sublayer may be included in the case of the most common
100baseTX version of fast ethernet. This standard, like 10baseT ethernet supports hub-to-endstation cabling distances of up to 100 m and allows a fast ethernet hub station to negotiate
with a newly connected station to determine the optimal speed and technology of transmission.
Auto-negotiation is conducted by means of a fast link pulse (FLP) exchanged between the two
devices when they are first switched on or connected. Alternatively, some hubs also support
autosensing. Autosensing allows older 10baseT ethernet devices to be incorporated into a
100baseTX fast ethernet network. The combination of devices is possible because the hub
Described later in this chapter.
Alternative physical layers — ethernet, fast ethernet and gigabit ethernet
is capable of autosensing the actual maximum bit rate of the end station and adjusting its
mode of operation accordingly. When provided with autosensing, the 100baseTX form of fast
ethernet is often referred to as 10/100baseT. This is becoming the most common form of fast
ethernet for connecting end stations to the hub.
Most fast ethernets and 10/100baseT networks are operated as a switched star topology
rather than as a collision domain. This is because an active hub in a collision domain has to
‘throttle back’ the rate of data transmission from the devices connected to the higher speed
ports, so that the frames can be broadcast even to the slower speed ports. By doing so, the
100 Mbit/s ports are effectively reduced to 10 Mbit/s throughput — even when communicating
with another 100 Mbit/s port. Collision domains make sense only in LANs which comprise
exclusively either 10 Mbit/s or 100 Mbit/s devices, otherwise switches and point-to-point connection of the end-devices makes more sense.
The 100baseFX version of fast ethernet is most commonly found as a point-to-point full
duplex technology for interconnecting different LANs on large campus sites using optical
The most commonly used connectors for fast ethernet are RJ-45 (for copper cable) and the
MIC (medium interface connector) or SC-connector (for fibre cables).
Gigabit ethernet physical layer (1000baseX, IEEE 802.3x, 802.3z and 802.3ab)
The protocol sublayer model of Gigabit ethernet (1000baseX) is similar to that of fast ethernet
(100baseX), as is clear from Figure 4.9. The most striking immediate differences are the
dispensing with the auto negotiation (which maybe can return once the 1000baseT standard
stabilises) and the change of the MII to the GMII interface.
The GMII (Gigabit medium-independent interface) interface uses two 8-wire transmission
paths for both the transmit and receive directions and is capable of transferring an entire
byte all at once. It employs 8B/10B line coding. Naturally, the RS (reconciliation sublayer)
for Gigabit ethernet is also slightly different from that of fast ethernet, but the same basic
functions of converting the 1-bit-at-a-time MAC layer format to the 8-bit parallel format of
GMII remain. Unlike the MII and AUI, GMII is not an inter-device physical interface but an
electronic component level interface within a Gigabit ethernet network interface card.
As with fast ethernet, the PMA (physical medium attachment) sublayer of Gigabit ethernet converts the internal parallel format of GMII into the precise format used on the
physical medium, and presents it to the lower physical medium dependent (PMD) and mediumdependent layer (MDL) for adaption for the medium itself. As with fast ethernet, there are a
number of different optional alternative physical realisations of Gigabit ethernet.
The fibre means of physical transmission for Gigabit ethernet uses serial line coding techniques based on the fibre channel physical transmission (standards which pre-dated Gigabit
ethernet). These are based on 8B/10B block coding and sent in NRZI (non-return-to-zero
inverted) line code. The copper cable alternatives meanwhile use high bandwidth coaxial
cable (1000baseCX) or multilevel transmission and multiple transmission leads (1000baseT):
• 1000baseLX, for a pair of multimode or monomode cables operating at 1300 nm wavelength;
• 1000baseSX, for a pair of multimode or monomode cables operating at 850 nm wavelength;
• 1000baseZX, for a pair of multimode or monomode cables operating at 1550 nm wavelength;
• 1000baseCX, for shielded balanced copper cable (twinax cable and connectors); and
• 1000baseT (IEEE 802.3ab) for 4-pair unshielded twisted pair category 5 cabling employing
the dual duplex transmission mode and PAM5 (5-state pulse amplitude modulation) coding
Local area networks (LANs)
All of the Gigabit ethernet interfaces are complex and expensive. Given this, and the fact that
many existing computers may be unable to handle the bit rates it makes possible, its use is
likely to be restricted to backbone and carrier networks for some time.
A note on fast ethernet and gigabit ethernet block codes and line codes
We discussed in Chapter 2 some of the most important line codes which are used to carry
digital data streams over a physical medium. We covered in detail the line codes NRZ (nonreturn-to-zero), NRZI (non-return-to-zero inverted) and Manchester coding. But you may now
be wondering about how the other line codes and block codes we have encountered in fast
ethernet and Gigabit ethernet operate:
• MLT-3 (multilevel transmission-3)
• 4D-PAM5 (pulse amplitude modulation-5)
For detailed documentation of the coding, you will need to refer to a databook or the relevant
specification, but it is nonetheless important to understand the principles:
4B/5B and 8B/10B are so-called block codes and not line codes. The block code 4B/5B
converts 4 bits of the original data stream (4B) into 5 bits of block code (5B) which are
actually transmitted on the line. Such block codes are used widely in conjunction with optical
line systems. They take advantage of the high bandwidth capability of optical fibre transmission
to increase the reliability of the system in terms of its synchronization and resistance to errors.
Thus a 4B/5B code increases the bit rate which needs to be carried by the line by 20%. But
by so doing, there is potential to ensure that repeat patterns of bits (e.g., 00000 or 11111) do
not lead to constant ‘on’ or ‘off’ signals on the line. The optical line coding itself can only be
two-state ‘on’ or ‘off’, and typically the NRZI line code is employed to transfer the 4B/5B or
8B/10B block-coded signal.
The 8B/10B code is also a block code which converts an entire byte of 8 bits into a block
codeword of 10 bit length.
8B/6T is a line code. It converts an 8-bit (8B) signal pattern into 6 digits of ternary (i.e.,
3-state) code (6T). An 8-bit pattern may have any one of 256 different binary values, as we
learned in Chapter 2. But a ternary (i.e., 3-state code, e.g., using three signal values, correctly
called symbols, +, 0 and −) can represent 36 = 729 different values using only 6 digits. By
using this line code we have ‘spare states’ [729 − 256 = 473 of them!] which we can use to
build in a capability equivalent to block coding, thereby making our signal less susceptible to
noise from other disturbing sources of electromagnetic interference (EMI). Meanwhile, the fact
that the line only has to carry 6 tertiary symbols rather than 8 bits means we have reduced
the baud rate. We can now get away with using a lower bandwidth medium! 8B/6T is a
specific form of 3-state multilevel transmission (MLT-3). Other well-known MLT-3 codes are
4B/3T (used across the ISDN U-interface), MMS43 and that defined by ANSI X3T9.5 (used
for 100 Mbit/s transmission over twisted pair cable).
4D-PAM5 is a 5-state pulse amplitude modulation transmission which allows an 8-bit signal
to be block-coded into 4 line-coded digits (each represented by one of the 5 different symbols
of the code, each symbol in turn being represented by a different line code pulse amplitude
allowed by the code).
LAN segments and repeaters — extending the size of a single collision domain
4.8 LAN segments and repeaters — extending the size of a single
Figure 4.10 illustrates three typical ethernet networks, showing how the number of devices
within a single ethernet LAN (in the jargon called a single collision domain) can be increased
by cascading end devices and/or hubs to create the necessary number of ports, provided the
cabling length does not exceed the maximum specified (Table 4.5).
The maximum geographic dimension7 of an ethernet LAN is determined as the product of
the maximum allowed cabling length within a segment multiplied by the maximum allowed
number of segments. A segment is a zone or subnetwork part of a collision domain. A segment
of a coaxial cable LAN (10base5 or 10base2) is that part of an ethernet LAN within which
end-users devices can intercommunicate with one another without passing a repeater. The
boundary of two different adjoining segments is thus established by means of a repeater.
In the days of coaxial cable-based ethernet (10base5 and 10base2), a repeater was necessary
to amplify the signals on the ethernet bus, in order that the signal would reach all segments
of the bus with sufficient strength to be correctly received by all connected stations (see
Figure 4.10c). One could determine how many segments a particular LAN comprised by
adding up the number of repeaters and adding one. Thus a network with no repeaters comprised
one segment. A LAN comprised of a ‘chain’ of subnetworks and two repeaters has three
segments, and so on.
In 10baseT networks, by comparison, repeaters tend to be referred to as hubs or switches,
and all the end-user ports attached to a single hub are considered to be in the same segment. A
Figure 4.10 Cascading segments and repeaters in 10Mbit/s ethernet LANs.
The maximum geographic dimension of a LAN is limited, as we saw earlier, by the maximum allowed
round-trip delay time.
Max 1 repeater
Max 1 repeater
320 m (HDX)
FDX) 3 km
Ethernet network size limitations
Max 1 repeater
HDX) 320 m
HDX) 550 m
Local area networks (LANs)