Practical TCP/IP and Ethernet Networking- P8 ppsx


6XGIZOIGR:)6/6GTJ+ZNKXTKZ4KZ]UXQOTM


integral number of octets and is either too short or too long. Provided all is correct, the
frame is passed to the LLC layer for further processing.
 )URROYOUTY
You should recognize that collisions are a normal part of a CSMA/CD network. The
monitoring and detection of collisions is the method by which a node ensures unique
access to the shared medium. It is only a problem when there are excessive collisions.
This reduces the available bandwidth of the cable and slows the system down while
retransmission attempts occur. There are many reasons for excessive collisions and you
will investigate these shortly.
The principle of collision cause and detection is shown in the following diagram.


Figure 3.7

CSMA/CD collisions

Assume that both node 1 and node 2 are in listen mode and node 1 has frames queued
to transmit. All previous traffic on the medium has ceased i.e. there is no carrier, and the
interframe gap from the last transmission has timed out. Node 1 now commences to
transmit its preamble signal, which immediately commences to propagate both left and
right on the cable. At the left end, the termination resistance absorbs the transmission, but
the signal continues to propagate to the right. However, the MAC sub layer in node 2 has
also been given a frame to transmit from the LLC sub layer, and since the node ‘sees’ a
free cable, it too commences to transmit its preamble. Again, the signals propagate on to
+ZNKXTKZTKZ]UXQY



the cable, and some short time later they ‘collide’. Almost immediately, node 2’s
transceiver recognizes that the signals on the cable are corrupted, and the logic
incorporated on the NIC asserts a collision detect signal. This causes node 2 to send a jam
signal of 32 bits of random data, and then stop transmitting. In fact, the standard allows
any data to be sent as long as, by design, it is not the value of the CRC field of the frame.
It appears that most nodes will send the next 32 bits of the data frame as a jam, since that
is instantly available.
This jam signal continues to propagate along the cable, as a contention signal since it is
‘mixed’ with the signal still being transmitted from node 1. Eventually, node 1 recognizes
the collision, and goes through the same jam process as node 2. You can see from this
that the frame from node 1 must be at least twice the end to end propagation delay of the
network, or else the collision detection will not work correctly. The jam signal from node
1 will continue to propagate across the network until absorbed at the far end terminator,
meaning that the system vulnerable period is three times the end to end propagation
delay.
After the jam sequence has been sent, the transmission is halted. The node then
schedules a retransmission attempt after a random delay controlled by a process known as
the truncated binary exponential back off algorithm. The length of the delay is chosen so
that it is a compromise between reducing the probability of another collision and delaying
the retransmission for an unacceptable length of time. The delay is always an integer
multiple of the slot time. In the first attempt, the node will choose, at random, either one
or zero slot times delay. If another collision occurs, the delay will be chosen at random
from 0, 1, 2 or 3 slot times, thus reducing the probability that a further collision will
occur. This process can continue for up to 10 attempts, with a doubling of the range of
slot times available for the node to delay transmission at each attempt. After ten attempts,
the node will attempt 6 more retries, but the slot times available for the delay period will
remain as they were at the tenth attempt. After 16 attempts, it is likely that there is a
problem on the network and the node will cease attempting to retransmit.
 3')LXGSKLUXSGZ
The basic frame format for an 802.3 network is shown below. There are eight fields in

each frame, and they will be described in detail.


Figure 3.8
MAC frame format

6XGIZOIGR:)6/6GTJ+ZNKXTKZ4KZ]UXQOTM


6XKGSHRK
This field consists of 7 octets of the data pattern 10101010. The preamble is used by the
receiver to synchronize its clock to the transmitter.
9ZGXZLXGSKJKROSOZKX
This single octet field consists of the data 10101011. It enables the receiver to recognize
the commencement of the address fields.
9U[XIKGTJJKYZOTGZOUTGJJXKYY
These are the physical addresses of both the source and destination nodes. The fields can
be 2 or 6 octets long, although the six-octet standard is the most common. The six-octet
field is split into two three octet blocks. The first three octets describe the block number
to which all NICs of this type belong. This number is the license number and all cards
made by this company have the same number. The second block refers to the device
identifier, and each card will have a unique address under the terms of the license to
manufacture. This means there are 248 unique addresses for Ethernet cards.
There are three addressing modes that are available. These are:
• Broadcast
Destination address is set to all 1s or FFFFFFFFFFFF
• Multicast
First bit of the destination address is set to a 1. It provides group restricted
communications
• Individual, or point-to-point

First bit of the address set to 0, and the rest set according to the target
destination node
2KTMZN
A two-octet field that contains the length of the data field. This is necessary since there is
no end delimiter in the frame.
/TLUXSGZOUT
The information that has been handed down from the LLC sub layer.
6GJ
Since there is a minimum length of frame of 64 octets (512 bits or 576 bits if the
preamble is included) that must be transmitted to ensure that the collision mechanism
works, the pad field will pad out any frame that does not meet this minimum
specification. This pad, if incorporated, is normally random data. The CRC is calculated
over the data in the pad field. Once the CRC checks OK, the receiving node discards the
pad data, which it recognizes by the value in the length field.
,)9
A 32-bit CRC value that is computed in hardware at the transmitter and appended to the
frame. It is the same algorithm used in the 802.4 and 802.5 standard.


+ZNKXTKZTKZ]UXQY


 *OLLKXKTIKHKZ]KKTGTJ+ZNKXTKZ
As has already been discussed, there is a difference between an 802.3 network and a Blue
Book Ethernet network. These differences are primarily in the frame structure and are
tabulated below.


Table 3.1
Differences between IEEE 802.3 and Blue Book Ethernet (V2)

The significant difference in the frame is the length field in 802.3 is interpreted as the
higher protocol field in Ethernet. Since an 802.3 frame cannot be longer than 1500 bytes,
the values in the protocol type field of the Ethernet V2 frame commences at 1500. This
allows protocol analyzers to recognize one type of frame as opposed to the other.
 8KJ[IOTMIURROYOUTY
The main reasons for collision rates on an Ethernet network are:
• The number of packets per second
• The signal propagation delay between transmitting nodes
• The number of stations initiating packets
• The bandwidth utilization

A few suggestions on reducing collisions in an Ethernet network are:
• Keep all cables as short as possible
• Keep all high activity sources and their destinations as close as possible.
Possibly isolate these nodes from the main network backbone with
bridges/routers to reduce backbone traffic
• Use buffered repeaters rather than bit repeaters
• Check for unnecessary broadcast packets that are aimed at non existent
nodes
• Remember that the monitoring equipment to check out network traffic can
contribute to the traffic (and the collision rate)
 +ZNKXTKZJKYOMTX[RKY
The following design rules on length of cable segment, node placement and hardware
usage should be strictly observed.


6XGIZOIGR:)6/6GTJ+ZNKXTKZ4KZ]UXQOTM


 2KTMZNULZNKIGHRKYKMSKTZ

It is important to maintain the overall Ethernet requirements as far as length of the cable
is concerned. Each segment has a particular maximum length allowable. For example,
10Base2 allows 200 m maximum length. The recommended maximum length is 80% of
this figure. Some manufacturers advise that you can disregard this limit with their
equipment. This can be a risky strategy and should be carefully considered.

Table 3.2
Length of the cable segment
Cable segments need not be made from a single homogenous length of cable, and may
comprise multiple lengths joined by coaxial connectors (two male plugs and a connector
barrel). Although Thicknet (10Base5) and Thinnet (10Base2) cables have the same
nominal 50-ohm impedance they can only be mixed within the same 10Base2 cable
segment to achieve greater segment length.
To achieve maximum performance on 10Base5 cable segments, it is preferable that the
total segment be made from one length of cable or from sections off the same drum of
cable. If multiple sections of cable from different manufacturers are used, then these
should be standard lengths of 23.4 m, 70.2 m or 117 m (± 0.5 m), which are odd multiples
of 23.4 m (half wavelength in the cable at 5 MHz). These lengths ensure that reflections
from the cable-to-cable impedance discontinuities are unlikely to add in phase. Using
these lengths exclusively a mix of cable sections should be able to be made up to the full
500 m segment length.
If the cable is from different manufacturers and you suspect potential mismatch
problems, you should check that signal reflections due to impedance mismatches do not
exceed 7% of the incident wave.
 3G^OS[SZXGTYIKO\KXIGHRKRKTMZN
In 10Base5 systems the maximum length of the transceiver cables is 50 m but it should be
noted that this only applies to specified IEEE 802.3 compliant cables. Other AUI cables
using ribbon or office grade cables can only be used for short distances (less than 12.5 m)
so check the manufacturer specifications for these!
 4UJKVRGIKSKTZX[RKY

Connection of the transceiver media access units (MAU) to the cable causes signal
reflections due to their bridging impedance. Placement of the MAUs must therefore be
controlled to ensure that reflections from them do not significantly add in phase.
In 10Base5 systems the MAUs are spaced at 2.5 m multiples, coinciding with the cable
markings.
In 10Base2 systems the minimum MAU spacing is 0.5 m.

+ZNKXTKZTKZ]UXQY


 3G^OS[SZXGTYSOYYOUTVGZN
The maximum transmission path is made of five segments connected by four repeaters.
The total number of segments can be made up of a maximum of three coax segments
containing station nodes and two link segments, having no intermediate nodes. This is
summarized as the 5-4-3-2 rule. These link segments are 10BaseFL fiber links as
specified in IEEE 802.3.


Table 3.3
5-4-3-2 rule
It is important to verify that the above transmission rules are met by all paths between
any two nodes on the network.


Figure 3.9
Maximum transmission path
Note that the maximum sized network of four repeaters supported by IEEE 802.3 can
be susceptible to timing problems. The maximum configuration is limited by propagation
delay.
Note that 10Base2 segments should not be used to link 10Base5 segments.

 3G^OS[STKZ]UXQYO`K
10Base5 = 2800 m node to node
(5 × 500 m segments + 4 repeater cables + 2 AUI)
10Base2 = 925 m node to node (5 × 185 m segments)
10BaseT = 100 m node to hub
 8KVKGZKXX[RKY
Repeaters are connected to transceivers that count as one node on the segments.
Special transceivers are used to connect repeaters and these do not implement the signal
quality error test (SQE).
Fiber optic repeaters are available giving up to 3000 m links at 10 Mbps. Check the
vendor’s specifications for adherence with IEEE 802.3 repeater performance and
compliance with the fiber optic inter repeater link (FOIRL) standard.


6XGIZOIGR:)6/6GTJ+ZNKXTKZ4KZ]UXQOTM


 )GHRKY_YZKSMXU[TJOTM
Grounding has safety and noise connotations. IEEE 802.3 states that the shield conductor
of each coaxial cable shall make electrical contact with an effective earth reference at one
point only.
The single point earth reference for an Ethernet system is usually located at one of the
terminators. Most terminators for Ethernet have a screw terminal to which a ground lug
can be attached using a braided cable preferably to ensure good earthing.
Ensure that all other splices taps or terminators are jacketed so that no contact can be
made with any metal objects. Insulating boots or sleeves should be used on all in-line
coaxial connectors to avoid unintended earth contacts.
4
Fast and gigabit Ethernet systems
Objectives

When you have completed study of this chapter you should be able to:
• List the basic methods used to achieve high transmission speeds on UTP
cables
• Describe the operation of the 100Base-TX system
• List the different physical media options for 100Base-T systems
• Explain the basic differences between a Class I and Class II repeater
• Explain the packet bursting technique used by gigabit Ethernet
• List the different media options used by gigabit Ethernet
4.1 Achieving higher speed
Although Ethernet with over 200 million installed nodes world-wide is the most popular
method of linking computers on a network, its 10 Mbps speed is too slow for very data
intensive or real-time applications.
From a philosophical point of view there are several ways to increase speed on a
network. The easiest, conceptually, is to increase the bandwidth and allow faster changes
of the data signal. This requires a high bandwidth medium and generates a considerable
amount of high frequency electrical noise on copper cables, which is difficult to suppress.
The second approach is to move away from the serial transmission of data on one circuit
to a parallel method of transmitting over multiple circuits at each instant. A third
approach is to use data compression techniques to enable more than one bit to be
transferred for each electrical transition. A fourth approach used with 1000 Mbps gigabit
Ethernet is to operate circuits full-duplex, enabling simultaneous transmission in both
directions.
All of the three approaches are used to achieve 100 Mbps fast Ethernet and 1000 Mbps
gigabit Ethernet transmission on both fiber optic and copper cables using the current
high-speed LAN technologies.
60 Practical TCP/IP and Ethernet Networking
4.1.1 Cabling limitations
Typically most LAN systems use either coaxial cable, shielded (STP) or unshielded
twisted pair (UTP) or fiber optic cables. The capacitance of the coaxial cable imposes a
serious limit to the distance over which the higher frequencies that can be handled.

Consequently 100 Mbps systems do not use coaxial cables.
The unshielded twisted pair is obviously popular because of ease of installation and low
cost. This is the basis of the 10Base-T Ethernet standard. The category 3 cable enables us
to achieve only 10 Mbps while category 5 cables can attain 100 Mbps data rates, whilst
the four pairs in the standard cable enable several parallel data streams to be handled.
As we have seen fiber optic cables have enormous bandwidths and excellent noise
immunity so are the obvious choice for high-speed LAN systems.
4.2 100Base-T (100Base-TX, -T4, -FX, -T2)
This is the preferred approach to 100 Mbps transmission, which uses the existing Ethernet
MAC layer with various enhanced physical media dependent (PMD) layers to improve
the speed. These are described in the IEEE 802.3u and 802.3y standards as follow.
IEEE 802.3u defines three different versions based on the physical media:
• 100Base-TX which uses two pairs of category 5 UTP or STP
• 100Base-T4 which uses four pairs of wires of category 3, 4 or 5 UTP
• 100Base-FX which uses multimode or single-mode fiber optic cable

IEEE 802.3y:
• 100Base-T2 which uses two pairs of wires of category 3, 4 or 5 UTP


Figure 4.1

Summary of 100Base-T standards
This approach is possible because the original 802.3 specifications defined the MAC
layer independently of the various physical PMD layers it supports. As you will recall,
the MAC layer defines the format of the Ethernet frame and defines the operation of the
CSMA/CD access mechanism. The time dependent parameters are defined in the 802.3
specifications in terms of bit-time intervals and so is speed independent. The 10 Mbps
Ethernet interframe gap is actually defined as an absolute time interval of 9.60
microseconds, equivalent to 96 bit times; while the 100 Mbps system reduces this by ten

times to 960 nanoseconds.
One of the limitations of the 100Base-T systems is the size of the collision domain,
which is 250 m. This is the maximum sized network in which collisions can be detected;
being one tenth of the size of the maximum 10 Mbps network. This limits the distance
Fast and gigabit Ethernet systems 61
between our workstation and hub to 100 m, the same as for 10Base-T, but usually only
one hub is allowed in a collision domain. This means that networks larger than 200 m
must be logically connected together by store and forward type devices such as bridges,
routers or switches. However, this is not a bad thing, since it segregates the traffic within
each collision domain, reducing the number of collisions on the network. The use of
bridges and routers for traffic segregation, in this manner, is often done on industrial
CSMA/CD networks.
The dominant 100Base-T system is 100Base-TX, which accounts for about 95% of all
fast Ethernet shipments. The 100Base-T4 systems were developed to use four pairs of
category 3 cable; however few users had the spare pairs available and T4 systems are not
capable of full-duplex operation, so this system has not been widely used. The 100Base-
T2 system has not been marketed at this stage, however its underlying technology using
digital signal processing (DSP) techniques is used for the 1000Base-T systems on two
category 5 pairs. With category 3 cable diminishing in importance, it is not expected that
the 100Base-T2 systems will become significant.
4.2.1 IEEE 802.3u 100Base-T standards arrangement
The IEEE 802.3u standard fits into the OSI model as shown in Figure 4.2. You will note
that the unchanged IEEE 802.3 MAC layer sits beneath the LLC as the lower half of the
data link layer of the OSI model.
Its physical layer is divided into the following two sub layers and their associated
interfaces:
• PHY – physical medium independent layer
• MII – medium independent interface
• PMD – physical medium dependent layer
• MDI – medium dependent interface

A convergence sub layer is added for the 100Base-TX and -FX systems, which use the
ANSI X3T9.5 PMD layer which was developed for the reliable transmission of 100 Mbps
over the twisted pair version of FDDI. The FDDI PMD layer operates as a continuous
full-duplex 125 Mbps transmission system, so a convergence layer is needed to translate
this into the 100 Mbps half-duplex data bursts expected by the IEEE 802.3 MAC layer.

Figure 4.2
100Base-T standards architecture