Introduction to Computer Networking
Ethernet Technologies

This lecture is divided into hyperlinked sections

Introduction
Coaxial Versions of Legacy Ethernet
10BASE-T - Legacy using Twisted Pair Cabling
Commonality in Legacy Ethernet

Ethernet Frame Components

Digital Encoding of Digital Data
10 BASE-T goes Full-Duplex
100 Mbps Ethernet

4B/ 5B Encoding

100 BASE-TX

100 BASE-FX

Gigabit Ethernet (1000 BASE-X)

1000 BASE-T
1000 BASE-SX and 1000 BASE-LX

10 Gbps Ethernet

Conclusion


Introduction
 Ethernet and the IEEE 802.3 protocols have become the world's most widespread LAN technologies. One of the reasons for this is the ease of installation and the evolution of this technology as user needs have changed since its inception. As a consequence, the expertise required to install and maintain Ethernet LANs is widespread and this helps to maintain its market position.

In this lecture we shall examine the different Ethernet technologies available.

Firstly we shall examine older Ethernet technologies and see the common features between implementations
Then we will examine 100 Mbps Ethernet and see the changes necessary to implement this technology.
We will see the differences and similarities involved with Gigabit Ethernet and finally look at the implications of 10 Gbps Ethernet.



Coaxial Versions of Legacy Ethernet

We shall not dwell long on the older Ethernet technologies. These used coaxial cable as the physical layer and the cabling scheme in use was the bus topology. This meant that a length of coaxial cable would be laid around the area in which the LAN was to be installed and the LAN nodes would be directly connected to the cable. The signalling used in the early installations (10BASE5 and 10BASE2) was baseband i.e. digital meaning that only one signal could exist on the bus cabling at any one time. The distance that the LAN could cover (without repeaters to return the signals to full strength) dropped from 500m to 185m as Ethernet migrated from 10BASE5 to 10BASE2.

10BASE-T   Legacy using Twisted Pair Cabling

The next Ethernet technology available was 10BASE-T. This made a change from the coaxial medium used up to this point and began to use Unshielded Twisted Pair (UTP) copper cabling. Unfortunately, the maximum length (without repeaters) of a network segment became reduced to 100m and thus the need for repeaters (two port devices) and multiport repeaters (hubs) was born. This allowed 10BASE-T networks to increase their coverage.

UTP cabling is much easier to install then coaxial cabling due to its flexibility and ease of connection and soon many LANs were rewired to take advantage of this new technology. Although 10BASE-T could use Category 3 cabling, many LAN owners used Cat 5 cabling as their physical layer.

It is instructive to note that this was a good policy because when higher data rate Ethernet (100 Mbps) became available a few years later, it was just the NICs and hubs that would need replacing and the cabling could remain untouched. This made for a large cash saving whan migration occurred. In general, it is much more expensive to pay for the labour when a cable is laid than the cost of the cable itself!

As LANs became more popular, users began to take advantage of the power of a LAN and the data traffic increased on LANs. The broadcast nature of Ethernet became a serious limitation to its operation and a solution was sought. Salvation came in the form of the Layer 2 switch. This allowed point-to-point links between communicating devices and freed precious network bandwidth for the use of other user on the LAN, however
Layer 2 switches remained extremely expensive until the turn of the century.
Commonality in Legacy Ethernet

These three older technologies (10BASE5, 10BASE2, 10BASET) shared the 10 Mbps data rate and this allowed re-use of the:

Timing parameters;
Frame format;
Transmission process;
Basic Design Rules.

The table below shows some of the parameters associated with 10 Mbps Ethernet.

Parameter
Value
Bit time
100 nanoseconds
Slot time
512 bit times = 512 x 100 x 109 = 51.2 microseconds
Interframe spacing
96 bits = 9.6 microseconds
Collision attempt limit
16
Collision backoff limit
10
Collision jam size
32 bits = 3.2 microseconds
Maximum frame size
1518 bytes
Minimum frame size
64 bytes

Ethernet Frame Components

The size of an Ethernet frame is fixed to lie between 64 and 1518 bytes. The frame is a digital structure that carries the data for the user and control information for the network. This structure therefore carries addressing information, some control information and error checking information. Together, these three parts occupy 18 bytes of the frame.

Legacy Ethernet is an asynchronous technology. This means that frames may be transmitted at any time and therefore the recieving Network Interface Card (NIC) must be given some warning of the arrival of an incoming frame. When a frame is transmitted on a LAN, the receiving NIC synchronises with the arriving frame by listening for a 7 byte preamble consisting of alternating 1s and 0s followed by a 1 byte Start of Frame field. These 8 bytes are not considered part of the frame - they only serve to allow receiving NICs to synchronise with the arriving data.

Below is shown the structure of an Ethernet frame for 10 Mbps Ethernet.

Preamble - 7 bytes
Start of Frame Delimiter - 1 byte
Destination -6 bytes
Source - 6 bytes
Length/ Type - 2 bytes
Data - 46 - 1500 bytes
FCS - 4 bytes

 Shown in yellow above are the sections that compromise the frame. The first two sections, coloured blue, are required for synchronisation when the frame is being received.

The data that makes up the frame must be converted to electrical signals so that it can be transmitted on the network cabling. The line encoding used is called Manchester Encoding and also manages to carry clocking information to the recieving NIC.

Digital Encoding of Digital Data

The most common, and easiest, way to transmit digital signals is to use baseband signaling to provide two different voltage levels for the two binary digits. Typically, a negative voltage is used to represent binary one and a positive voltage is used to represent binary zero. This code is known as Nonreturn-to-Zero-Level (NRZ-L) meaning the signal never returns to zero voltage, and the value during a bit time is kept level  being either a negative or a positive voltage.

A significant disadvantage of NRZ transmission is that it is difficult for the receiver to determine where one bit ends and another begins. To picture the problem, consider that with a long string of ones or zeros for NRZ-L, the output is a constant voltage over a long period. Under these circumstances, any drift between the timing of transmitter and receiver results in the loss of synchronisation between the two. 

As such, this is unsuitable for an Ethernet network. A signaling method that carries timing information is required to help solve this. Manchester encoding is used in Ethernet networks to carry the signals across the wiring.

Figure depicting various methods available for Digital Signal Encoding

Manchester encoding overcomes the problem of timing with long strings of 1s and 0s, using a set of alternative coding techniques, grouped under the term biphase. Two of these techniques, Manchester, (used in Ethernet) and Differential Manchester, are in common use.

All the biphase techniques require at least one transition per bit time and may have as many as two transitions. Thus, the maximum modulation rate is twice that for NRZ; this means that the bandwidth required is correspondingly greater. To compensate for this, the biphase schemes have the following advantages:

Synchronisation: Because there is a predictable transition during each bit time, the receiver can synchronise on that transition. For this reason, the biphase codes are known as self-clocking codes.

Error detection: The absence of an expected transition can be used to detect errors. Noise on the line would have to invert both the signal before and after the expected transition to cause an undetected error.

Ethernet uses the Manchester encoding system. In the Manchester code, there is a transition at the middle of each bit period. The mid-bit transition serves as a clocking mechanism and also as data: A high-to-low transition represents a zero, and a low-to-high transition represents a one.

In Differential Manchester, the mid-bit transition is used only to provide clocking. The encoding of a zero is represented by the presence of a transition at the beginning of a bit period, and a one is represented by the absence of a transition at the beginning of a bit period.

Differential Manchester has the added advantage of employing differential encoding. In differential encoding, the signal is decoded by comparing the polarity one signal element with its preceding element rather than determining the absolute value of a signal element. One benefit of this scheme is that it may be more reliable to detect a transition in the presence of noise than to compare a value to a threshold. Another benefit is that with a complex transmission layout, it is easy to lose the sense of the polarity of the signal. For example, on a multipoint twisted-pair line, if the leads from an attached device to the twisted pair are accidentally inverted, all ones and zeros for NRZ-L will be inverted and the sense of the message will be lost. This cannot happen with differential encoding.



10 BASE-T goes Full-Duplex
In 10 BASE-T networks, all cables are plugged into a hub (multiport repeater) or a switch. The hub contains the circuitry to simulate the shared medium and thus it is as if the bus cabling from the coaxial implementations of Ethernet were collapsed into the hub itself.

Originally, 10 BASE-T was specified as a half-duplex network, however it was later redesigned to allow for full-duplex operation. It is an asynchronous network meaning that frames may arrive at any time and therefore there may be long inter-frame periods during which no transmissions take place. To fill this silence on the network cabling, link pulses are released every 125 microseconds.

On the cabling are four pairs of wires. Only two pairs are used for communication in 10 BASE-T. One pair is used for transmission and another pair is used for receiving signals. This means that no collisions can take place on the network cabling because there are two separate transmission paths. When used in conjunction with a hub, the collision of two frames from different stations will not occur in the network cabling but within the hub itself. Therefore it is only possible to run a network in half duplex when using a hub.

Hubs are subject to the 5-4-3-2-1 rule that states that there can only be a maximum of 5 network segments separated by 4 hubs. This limits the range of coverage of the network. The rule exists to prevent violation of the timing rules that are present to prevent collisions of frames. This limitation can be overcome by replacing all hubs with layer 2 switches. By using switches, the network can be extended and the 5-4-3-2-1 rule can be ignored. Circuitry within the switch allows the switch to deal with timing issues and collisions.


Switches may be daisy chained and so long as the 100m cable length is not violated, the network may be extended as much as required.  

100 Mbps Ethernet
This is often known as Fast Ethernet. There are two implementations, a copper-based physical layer (100BASE-TX) and a fibre based physical layer (100BASE-FX). The 100BASE-TX solution requires the use of Cat 5 (or better) cabling and the 100BASE-FX solution requires the use of multimode fibre optic cabling.  Both implementations share the same frame format, timing parameters and transmission process.

The table below shows parameters for 100 Mbps Ethernet
Parameter
Value
Bit time
10 nanoseconds
Slot time
512 bit times = 512 x 10 x 109 = 5.12 microseconds
Interframe spacing
96 bits = 0.96 microseconds = 960 nanoseconds
Collision attempt limit
16
Collision backoff limit
10
Collision jam size
32 bits = 0.32 microseconds = 320 nanoseconds
Maximum untagged frame size
1518 bytes
Minimum frame size
64 bytes

If we compare this to the previous table, it can be seen that the bit time is 1/10 th of that for 10 Mbps Ethernet. All other parameters that depend on the timing are reduced correspondingly by 1/10th also, however the basis for the technology remains the same.

Because the data rate is 10 times that of 10 Mbps Ethernet, the bits arrive 10 times faster. This approaches the frequency limit of Category 5 cabling. Manchester encoding is no longer suitable here and a different technology has to be used to overcome new issues raised by the increase of data rate. These issues are synchronisation and signal to noise ratio (SNR). At 100 Mbps, the SNR becomes more of an issue because a momentary spike has a 10 fold increase in duration compared to 10 Mbps implementations.

4B/ 5B Encoding
This encoding scheme takes nibbles of data (4 bits) and converts them to 5 bit representations.

Each group of 4 bits is encoded into a 5-bit (binary) symbols to guarantee that the transmitted signals will contain no more than two consecutive 0s. This prevents errors occurring due to timing drift at the receiver.

The table below gives a list of 4B/5B Codes for frame data. Other 5-bit groups are also used to send control information across the network (not shown here). The other 5 bit groups are used to convey control information such as start of frame, end of frame or to indicate that the line is idle. One symbol is transmitted during idle conditions. This symbol allows autosensing equipment to detect whether the network it is connected to is 10 or 100 Mbps. These are known as Fast Link Pulses.
 
4-bit vs. 5-bit group 4-bit vs. 5-bit group
0000
0001
0010
0011
0100
0101
0110
0111
11110
01001
10100
10101
01010
01011
01110
01111
1000
1001
1010
1011
1100
1101
1110
1111
10010
10011
10110
10111
11010
11011
11100
11101

 Every time 4 bits of data are  generated by the user, they are converted to a 5 bit symbol. This means that to transmit 100 Mbits, 125 Mbits need to be transmitted across the physical medium. More careful testing of the physical layer is required to ensure that the cabling is suitable to carry such frequencies.

Now we have bits to transmit, it is necessary to place tham onto the network cabling. This differs whether copper or fibre optic cabling is used.

100 BASE-TX
In copper based Fast Ethernet, switches must be used instead of hubs if full duplex operation is required. Dual speed switches (10/ 100 Mbps) were introduced to allow for older legacy 10 Mbps equipment to be used alongside the faster 100 Mbps equipment.

As there are two separate data paths, no collisions will occur on the cabling and collisions will be mitigated in the switches. The network may be configured for half or full duplex operation. When using switches, all connections will be point-to-point and collisions are administratively detected for half duplex operation (even though they do not occur!) and the rules of CSMA/ CD must be used. In full duplex operation, both parties in the point-to-point link may transmit simultaneously raising the data rate to 200 Mbps. Most of the time it is wise to set Fast Ethernet to operate in full duplex mode.

To place the signal on the cabling, an encoding method known as MLT-3 is used. This encoding scheme uses THREE voltage levels to convey data across the physical layer  (+5v, 0v, -5v). To represent a 1, there is a voltage transition; to represent a 0 there is no transition. Hence the occurrence of repeated 1s will provide a transition (preventing timing drift) and a the maximum number of 0s that can appear together (see 4b/ 5B table above) is two, therefore timing drift at the receiver is unlikely over 2 bit times.

The figure below depicts a bit stream encoded using MLT-3 encoding. If there is a mid-bit transition, it indicates the presence of a 1, otherwise the voltage remains the same indicating a 0.


100 BASE-FX
The fibre option for Fast Ethernet was created to allow for vertical cabling and backbone applications. It is also suitable for high-noise environments or areas where electricity is undesirable (e.g. where explosive materials are present).

100 BASE-FX uses 4B/ 5B to encode the data but for line encoding uses NRZI line encoding. Fibre links operate by sending pulses of light at either high or low (almost no light) intensity. NRZI indicates a 1 with a mid-bit transition (either high or low) and a 0 by no transition. The figure below shows the pattern generated for a particular bit pattern.


When using fibre, one strand is used for transmission of data and another is used for reception of data.

Gigabit Ethernet (1000 BASE-X)
This version of Ethernet transmits 10 times faster then Fast Ethernet. The frame format, MAC addressing and CSMA/ CD are retained, however there are changes to the MAC sublayer, and the physical layer. Gigabit is designed to operate in full duplex mode. Although designed for fibre, it is possible to run Gigabit over UTP (1000 BASE-T).

The table below shows the parameters for Gigabit Ethernet.

Parameter
Value
Bit time
1 nanosecond
Slot time
4096 bit times = 4096 x 1 x 109 = 4.096 microseconds
Interframe spacing
96 bits = 96 nanoseconds
Collision attempt limit
16
Collision backoff limit
10
Collision jam size
32 bits = 32 nanoseconds
Maximum frame size
1518 bytes
Minimum frame size
64 bytes
Burst limit
65 536 bits = 8192 bytes
 
As the data rate is even higher now, encoding is needed to overcome noise and timing issues. On 1000 BASE-T networks, the encoding used is called 8Bit-1 Quinary Quarter (8B1Q4) and the line encoding (for copper) is known as 4-Dimensional 5 Level Pulse Amplitude Modulation (4D-PAM5). For UTP operation, all 4 pairs of cable are used in parallel. In this technology, the cables are used in full duplex mode which requires complex decoding of arriving signals because both transmitted and received signals will be present on the wires.

1000 BASE-T
This runs Gigabit over UTP and needs to use all 4 pairs simultaneously to manage such a high data rate. Signals to be transmitted are split into 4 separate paths and transmitted over the 4 pairs within the cabling. This means that the data rate is 250 Mbps per pair of cables. The 4 signal paths are recombined on arrival at the destination NIC.

Full duplex operation means that there are constant collisions on the wire pairs, however for Gigabit, this is perfectly permissible. Complex circuitry in the NICs use various techniques to separate the transmitted and received signals.
One technique known as echo cancellation assists in reception of signals. Another technique known as Forward Error Correction (FEC) helps achieve Gigabit data rates.

This complex system relies on the adherence to the cabling, termination and noise guidelines. 

1000 BASE-SX and 1000 BASE-LX 
These technologies are both fibre based. 1000 BASE-SX uses short wavelength light (850 nm) and operates over multimode fibre. 1000 BASE-LX uses long wavelenegth light and operates over single mode fibre. An encoding scheme known as 8B/ 10B is used and the line encoding is NRZ. Here, a low level of light indicates a 0 and a high level of light indicates a 1. Two fibres are used, one for transmission and the other for reception.

10 Gbps Ethernet
This was introduced during 2002 as IEEE 802.3ae-2002. It offers suitability for LANs and also MANs and WANs. The frame format is still the same as earlier versions. The physical layer standards have changed to allow a maximum distance of 40 km over single mode fibre. 10 Gbps Ethernet (10 GE) is also compatible with SONET/ SDH networks and ATM.

All implementations of 10 Gbps Ethernet run over fibre cable, different implementations suiting different applications. This is full duplex only. CSMA/ CD has finally been removed from this Ethernet specification. When 10 Gbps Ethernet is run over a MAN or a WAN, it is likely that the amount of data that is carried in a frame will be increased to take advantage of the low error rates associated with fibre. This will finally break the original frame format that can only carry 1500 bytes of user data.

Conclusion

There are many common parameters between all implementations of 10 Mbps Ethernet and the frame format used.
The frame format does not change across all Ethernet implementations up to 10 Gbps Ethernet.
Manchester line encoding is used at 10 Mbps to
carry timing information for the benefit of the receiver.
At 100 Mbps, a technique known as 4B/ 5B is used to provide extra symbols for signalling across the network.
Different techniques for line encoding are used depending whether copper or fibre is used as the physical layer.
At Gigabit speeds, all 4 pairs are used in full duplex mode to achieve the data rate.
Complex techniques are used to recover the data from the cabling.
The line encoding varies once again depending on the physical layer used.
10 Gbps Ethernet allows Ethernet to be used across MANs and WANs.

(c) MM Clements 2010