Local Area Networks

bluegooseexchangeΔίκτυα και Επικοινωνίες

26 Οκτ 2013 (πριν από 4 χρόνια και 13 μέρες)

124 εμφανίσεις

Local Area Networks

1




Local Area Networks


Brian Bramer

Department of Computing Sciences

DeMontfort University

Leicester UK




1 LAN Characteristics

................................
................................
................................
.........................

2

2 LAN topologies

................................
................................
................................
................................
.

3

3 MAC and LLC protocols

................................
................................
................................
..................

3

4 Topologies and MAC protocols
................................
................................
................................
......

4

4.1 Star configuration
................................
................................
................................
......................

4

4.2 Bus configuration (e.g. Ethernet, IEEE 802.3)

................................
................................
.....

5

4.2.1 Bus MAC (Medium Access Control) protocols

................................
................................
.

5

4.2.1.1 Carrier Sense Multiple Access (CSMA) protocol

................................
......................

5

4.2.1.2

Carrier
Sense Multiple Access/Collision Detect (CSMA/CD) protocol

..................

6

4.1.2.3

Throughput on CSMA/CD networks
................................
................................
............

7

4.1.2.4

IEEE 802.3

CSMA/CD bus
standards

................................
................................
........

8

4.1.2.5 Token passi ng bus
................................
................................
................................
.........

9

4.3 Ri ng configuration
................................
................................
................................
.....................

9

4.3.1

Ring MAC (Medium Access Control) Protocols
................................
..........................
10

4.3.1.1 Token passi ng ri ng (e.g. IEEE 802.5)
................................
................................
.......
10

4.3.1.2 FDDI (Fiber Distributed Data
Interface) Network (ANSI X3T9.5)
.........................
10

5 Hubs (or wiring centers)

................................
................................
................................
................
12

5.2 Intelligent Hubs

................................
................................
................................
.......................
12

6 Network Access: Probabilistic or Deterministic

................................
................................
.........
13

7 Internetworks: Bridges, Routers and Gateways
................................
................................
........
13

8 Distribu
ted Environments
................................
................................
................................
..............
15

8.1 Performance factors i n a distributed environment
................................
.............................
15

9 What network to buy?
................................
................................
................................
....................
16


Local Area Networks

2


Introduction


Local area networks have developed rapidly over recent years in response to the widespread distribution of
cheap stand
-
alone microcomputer systems. Users soon
realized

that the availability of cheap personal
distributed processing power o
n their desk was not sufficient. There is a need to share resources such as:

(a)

relatively expensive peripherals such as laser printers and tape backup devices,

(b)

information such as common data files and shared databases.

The latter being the most important
in
today’s

complex commercial and industrial
organizations
. A network
allows computer systems and computer based devices to communicate directly with each other over
a
common communications system,
e.g.
:



To connect terminals or personal computer systems t
o powerful
c
entralized

host computers.



To allow micro and minicomputers to access shared resources, e.g. high quality printers, large
disks (for sh
ared programs, databases, etc).



To transfer information between computers,
e.g.

programs, data, mail.


.

1
LAN

Characteristics

LANs have a number of common characteristics:



transmission medium is shared by all devices,
e.g.

connected by a common cable hence: transmission
by one device is received by all others,
i.e.

a
broadcast network



transmission is normally

in the form of packets (a message is split up into packets)



limited distribution of machines; up to 10 km and typically around 1km



connection of machines typically restricted to a single site,
e.g.

an industrial plant



high data rate; typically 1
0 times fa
ster than
WANs (wide area networks)



sharing of resources, e
.
g
.

users distributed around the site accessing common fileservers, printers,
plotters, etc., (WANs are usually used to transfer information between sites)



single ownership of all elements of the n
etwork; in particular communication lines are not owned by a
PTT
organization

such as BT



connection of incompatible equipment to the network,
i.e.

machines ranging from terminals, personal
microcomputers, large mainframes, etc. (from different manufacturer
s) running different software
(operating systems, file systems, network protocols, etc.).

Since

the communication lines are not owned by the PTT their bandwidth is not limited artificially and the error
incidence is lower. Thus much higher data rates can b
e maintained and the protocols can be simpler since
efficiency is not as crucial a consideration. Further, not using PTT lines automatically restricts the network to a
single private site and hence restricts the overall diameter of the network.







Local Area Networks

3

2 L
AN topologies



LANs are usually described in terms of their topology (logical layout
)
.



Star, bus and ring network topologies

Note that the topologies shown
above

are LOGICAL topologies
-

it does not necessar
ily follow that a LAN
which operates as a ring is physically wired with a single loop of cable. It is often inefficient to wire up LANs as
physical rings or buses (particularly in multi
-
story buildings).

Similarly bus networks are usually physically wired

as a
backbone spine

(e.g. vertically up a building) off which
individual cable segments are dropped by repeaters or bridges,
e.g.

on each floor. Often the shorter segments
are of lower specification but there are strict specifications on segment lengths,
number of stations per
segment, number of repeaters, etc.


3

MAC and LLC protocols

The lower three layers of the ISO 7 layer OSI reference model, Layers 1 (physical), 2 (data link) and 3
(network), are usually called the communications subnet (or just sub
net) separating the pure communications
aspects of the network from the application layers looked after by the stations,
i.e.
:

Physical Layer
is the physical node to node link (bit rate, signal levels, etc) which transfers raw data as bits
or bytes betwee
n adjacent nodes.

Data Link Layer
manages the node to node data transmission (flow control, error correction, etc) which
transfers data frames between adjacent nodes.

Network Layer
manages the network and the station to node link (the interface presented
by the network to
the station). Data transfer is in the form of packets.

The majority of LANs are broadcast networks in that all stations share a common communication medium and
receive the transmitted sign
al
. In this case the Data Link Layer is split into

two protocols:


MAC

Medium Access Control protocol
:


determines which device has access to the medium (cable) at
any time,
e.g.

CSMA/CD contention bus, token passing, etc.

LLC

Logical Link Control protocol
:

manages the link, i.e. flow control,

error correction (corrupt
frames, lost frames, multiple frames), etc.

Local Area Networks

4


Host computer connected to the subnet (lower three layers)

The above diagram

shows a pair of host computers connected to NIUs (Network Interface Uni
ts
-

probably a
plug
-
in card in the case of a PC) which handle the communication across the network.

In broadcast networks the MAC protocol also deals with physical NIU (Network Interface Unit) addressing,
i.e.

it removes the frame at the correct NIU and
passes it to the LLC protocol. Further addressing may be
performed in the LLC layer to identify particular devic
es attached to a NIU
. The LAN network layer manages
the host/NIU interface and splitting the host messages up into packets of a suitable size. I
n the case of a WAN
the connections can be very complex with many alternate paths between stations and addressing and routing
is a function of the Network layer.


4
Topologies

and MAC protocols



4.1

Star configuration

The star topology is best considered

as an internal telephone network in which the telephones are replaced by
computer devices and the PBX by a switching computer/interface unit. The operation is similar to that for
traditional wide area terminal networks. It was commonly used in microcomput
er networks in the early 1980's
when hard disks were very expensive; the machine at the centre of the star acting as a fileserver for the
others.


Today many networks are physically wired as stars using network hubs

(
see later

description of hubs)
althoug
h their logical structure may be bus or rings.
.

Local Area Networks

5

4.2

Bus configuration (
e.g.

Ethernet, IEEE 802.3)


Machines are connected to the network by physically tapping into a common cable (usually coaxial or twis
ted
pair). For example
.


Bus based network configuration (

are terminating resistors)

The stations are multi dropped off the cable with transmission by one station propagating in both directions
along the bus. All other stations

'hear' this transmission and the intended destination copies the data content.
The signals are absorbed by terminators at each end of the segment.

Most bus networks are based on the Ethernet specification developed by Xerox. The medium
was

usually
coaxia
l cable
which has been replaced over the past few years by twisted pair cable.

The bus is a passive mechanism; when no device is transmitting the cable is idle (no signal present) and if
any interface unit crashes the remainder of the network remains opera
tional. There is no need to remove
messages from the bus; after transmission they flow along the cable and are absorbed by the terminating
resistors at each end. However, there is a need for message receipt to be acknowledged explicitly by the
receiving de
vice,
e.g.

using a stop and wait data link protocol or better. Most bus networks are based on the
Ethernet specification and use a contention MAC protocol for sharing access to the cable, see next section.




4.2.1

Bus MAC (Medium Access Control) protocol
s

4.2.1.1

Carrier Sense Multiple Access (CSMA) protocol

Most bus protocols are based on a version of the contention protocol Carrier Sense Multiple Access (CSMA).
CSMA is an extension of a simple protocol developed at the University of Hawaii in the late 1
960s. The
problem with Hawaii is that it consists of a large number of islands which makes connection by cable very
difficult. It was decided instead to use radio but the designers were reluctant to allocate a fixed frequency
band to each device; radio ban
dwidth is a scarce resource and most of the time the devices would be idle.

They chose instead a very simple approach in which all devices used the same frequency and any device
would send its message whenever it felt like it. If only one device sent, the

signal would get through but if two
or more sent at the same time, the signals would
collide

to produce garbage. If the signal was not corrupted
the receiver would send an acknowledgement which, assuming no collision, would be received by the
transmitter.

This protocol is called ALLOHA (what else for Hawaii!) and is very inefficient. However, if traffic is
very low, this inefficiency doesn't matter. CSMA is a simple extension of this principle which improves the
efficiency.

The idea underlying Carrier Sen
se Multiple Access is that there is no point in transmitting if someone else is
already doing so. Therefore any station wishing to transmit first listens to the cable to see if it is busy. If it is
idle, the station sends its message otherwise it defers se
nding. How quickly it tries again is called the
persistence; it may:

Local Area Networks

6


a)

keep sensing for an idle cable continuously (1
-
persistent) or

b)

back off for a random amount of time before trying again (non
-
persistent) or

c)

generate a random number 0<r<1 and retry on
ly if r is greater than some number p (p
-
persistent).


Each time it finds the cable busy the random wait increases.

Sensing

that

the cable is idle does not guarantee that two stations will not send at the

same

time

causing

a

collision
, i.e. two signals cor
rupt each other. The

LLC

(
Logical

Link

Control
) protocol would
resend the frame if an acknowledgement of successful receipt has is not

been

received in a given time, i.e. it
will timeout and retransmit the frame.



4.2.1.2

Carrier

Sense Multiple Access/Co
llision Detect (CSMA/CD) protocol

Ethernet uses a simple extension of CSMA called Collision Detect (CSMA/CD) which is similar
in

philosophy

to

a party
-
line telephone system.


There is obviously no point in

continuing

to

send

if

a collision
has occurred; th
e message has already been destroyed.


In this

protocol

a

station

listens

to

its own
transmission and aborts the transmission if it hears a

collision
,

i.e.

what it 'hears' is not what it is transmitting.


The station transmits a short jamming signal so th
at

all

stations on the network will detect the collision and
then waits a random

amount

of

time before trying again.

In


practice Ethernet detects a collision by the interface monitoring the signal


level on the bus such that if the
signal exceeds the max
imum voltage swing that can be produced by a single


transmitter it assumes a
collision has occurred, i.e. the combination of the two signals which
collide
exceeding


the


maximum level of
one signal.


Attenuation (voltage loss as a signal

is

transmitted

d
own a cable) can present a potential problem
if

two

stations are far apart.


It both are transmitting (causing a collision) the signals may be so attenuated
that the combined signal does not exceed the detection threshold so neither station detects confli
ct.


This is
one reason why bus networks are restricted in length.


For example, Ethernet specifies a

maximum

segment
(individual bus cable) length of 500 meters which can be increased to a maximum overall network length of
2500 meters (5 segments) using u
p to a maximum of

4

repeaters

(
a

repeater

is

an amplifier which boosts the
signal and does nothing else; unlike bridges which also act as an address filter, see below).

This protocol has no merit unless the station can be certain to detect a collision bef
ore it

has

sent

much

of

the

message.


The time it takes to

be

certain

is called the collision window

and

is

one of the factors which limit the
maximum length of the bus.


Consider

two

stations A and B at the opposite ends of a coaxial cable 1km long.


In p
ractice an electrical signal takes approximately 5 microseconds to travel from one end of a
kilometer

coaxial cable to

the

other, i.e. 200 km/sec (electromagnetic waves such as light and radio waves travel at
2997925

km
/sec

in

vacuum

and are slower when
tr
aveling

in other material such as the dielectric of a coaxial
cable).

|

1.

A listens at time 0 and cable is free so it sends

2.

B


listens


at


time 4.9999 microseconds and thinks cable is


free


because


A's


signal
hasn't
arrived yet

3.

Collision occurs at time 5 microseconds

4.

A hears collision at time 10 microseconds and stops

Thus the longer the cable between stations the greater the length of time it will take to detect
collisions.


If


the


data


packet


is


small


and the


cable


is


too


long


A


may


have


finished
transmission of the packet before the collision occurs.

Assuming

that

signals

travel

at

a speed of 200 km/sec

the

collision

window

may

determined as
follows.


If a bus operates at R Mbit/sec and has a
maximum distance between stations


of K km, the
collision window will be equal to 10*K microseconds (the time it


takes a


signal to get to the other
end and back) or 10*K*R bits; thus the further or faster


you


go, the


longer


time


before


collisions


are known and the


more


message


will


be


wasted.


For example,

a

Ethernet

bus

operating at 10
Mbps over the maximum

length

of

2500

meters

would have a collision window of 25 microseconds
Local Area Networks

7

or 250 bits.

It


is


important


to


note


that even if a collisi
on does


not


occur


the


sending


station cannot assume
that the receiver has received the data OK, i.e. electrical noise on the line may have


corrupted


the
signal or the receiving station may be switched off.


If

received

OK the LLC (Logical Link Contr
ol)
protocol would provide an acknowledgement of receipt of the data.

4.1.2.3

Throughput

on CSMA/CD networks

The


protocol


outlined


above is called 1 persistent CSMA/CD in


which


stations


wishing


to transmit wait for
the line to become free then tran
smit
-

this has the problem that as


network traffic


increases


collisions


are


more frequent


and


unpredictable


performance


degradation occurs


when


the traffic reaches approximately
30% of the

bandwidth capacity
.

In non
-
persistent CSMA/CD the stat
ions wait a random amount of time before even looking to see

if

the

line

is
clear then apply random delays if it is busy.


This

results

in

improved

line

utilization

but at the cost of greater
average delays in particular at low line
utilization
.


CSMA
/CD protocol
-

offered load vers. net
utilization

The

use

of

intelligent hubs and bridges (see below) to break up a large network can help to alleviate the
problem.

Local Area Networks

8

4.1.2.4

IEEE

802.3

CSMA
/CD bus standards

The IEEE

802.3 committee has defined a number of alternative CSMA/CD standards
:

parameter

10BASE5

10BASE2

10BASET

100BASET

10BROAD36

Transmission
medium


coaxial
cable



(50 ohm)


coaxial
cable



(50 ohm)


Unshielded
twisted pair


Unshielded
twisted pai
r


coaxial
cable



(75 ohm)

Signaling

technique


Baseband



(Manchester)


Baseband



(Manchester)


Baseband



(Manchester)


Baseband



(Manchester)


Broadband



(DPSK)

data rate (Mbps)


10

10


10


100


10


maximum segment
length (meters)


500

185


100


100


1800



network span
(meters)


2500



(4 repeaters)


925



(4 repeaters)

500



(4 repeaters)

500



(4 repeaters)

3600



(1 repeater)


nodes per segment


100

100


-


-


100



cable diameter (mm)

10



Thick
Ethernet


5



Thin
Ethernet

0.4
-

0.6

0.4
-

0.6

0.4
-

1.0

The general notation for the implementations is:

<data rate in Mbits/s
ignaling

method <maximum segment length, 100's of meters

where the
signaling

method is BASE for baseband or BROAD for broadband.

The exception
to the length
is 10BASET

a
nd 1000BASET

where the T stands for twisted pair cable.



10BASE5

Was the original 802.3 standard based on the Ethernet bus using thick high quality (and
expensive) co
-
axial cable. Due to the cable being high
-
quality it has low attenuation and t
he
maximum segment length is 500 meters which can be extended to 2500 using a maximum of
4 repeaters (otherwise the
collision window

would become too large).

10BASE2

this is similar to 10BASE5 but uses thinner lower cost coaxial cable (sometimes called th
in
Ethernet or Cheapernet). Due to the lower quality cable the number of stations and the
segment length is reduced (but is usually adequate for small office networks).

10BASET

the T stands for unshielded twisted pair cable (cheap telephone cable) and, by

sacrificing
maximum segment length (100 meters), runs at a data rate of 10 Mbps (in the early 1980's
this tended replace 10BASE2 in many networks).

100BASET

is the latest addition to 802.3: runs at a data rate of 100 Mbps (this option is used where
netwo
rk load is high, e.g. systems using multimedia).

10BROAD36

is a broadband option providing support for more stations over a wider area but at greater
cost.


Local Area Networks

9

4.1.2.5

Token passing bus

In the token passing bus (e.g. IEEE 802.4) the stations are connected
physically by a common bus
but with the stations logically
organized

(by the network software) as a ring,
e.g.

ARCnet. A token
passing protocol (described below) is then used to control access to the bus.
The software of token
passing busses is very comple
x (handling such problems as nodes joining and leaving the network)
and they tend to be use for specialized applications where the advantages of both bus and ring are
required, e.g. factory automation.

4.3

Ring configuration

A ring network (e
.
g
.
IEEE 802.5
), as its name suggests, consists conceptually of a single loop of cable, usually
coaxial or twisted telephone pair, along which traffic f
lows in one direction
.


Ring based network configuration

Unlike the bus, the

cable passes through each Network Interface Unit, called a node or station, which repeats
the signal so that its strength is continuously maintained. Hence a ring can cover a larger distance than a bus
(where the transmitting station has to have sufficien
t power for the signal to reach the ends of the cable
without attenuation effecting collision detection). However, the ring is an active mechanism; if the cable is
broken or a node crashes, the ring would be disabled so most implementations contain duplica
te cable loops
and interface circuits,
i.e.

when a node is switched off a relay automatically connects the station ring input to
the ring output to maintain the integrity of the ring.

Since a message is continuously regenerated by each station it must be
explicitly removed (in a bus the
terminators absorb it). The convention is that the sender removes its message when it returns round the ring;
an acknowledgement of receipt can be 'piggybacked' onto the end of the original message by the receiver.
However,

interference on the ring could result in the transmitter not
recognizing

its message; the message
would then circulate endlessly unless a special station, the monitor, had responsibility for clearing garbage
messages.

There are two common ring MAC protoc
ols
-

the slotted ring which is the basis of the Cambridge Ring and the
Token Passing Ring which is the basis of the IBM token passing ring LAN product. The Ring was developed in
Europe and is moderately well established there; it is the basis of many chea
p microcomputer networks like
Econet and Clearway.

Local Area Networks

10

4.3.1

Ring MAC (Medium Access Control) Protocols

Bus configurations such as Ethernet using the CSMA/CD protocol are limited in length due to the time it takes
to detect a collision. Rings suffer from the

opposite problem in that they have to be long enough to contain the
bits of the message. Consider:

1.

The electrical signal travels at 200,000 km /sec

2.

The sender transmits at 1,000,000 bits /sec

Therefore in a 1 second message there would be 1,000,000 bits
spread over 200,000 km,
i.e.

each bit would
take up the equivalent of 200 meters of cable.

The minimum length of the ring must be sufficient to hold the token (see below). In practice the interface unit
of each station puts a one bit delay into the ring a
nd shift registers can be used to increase this.




4.3.1.1

Token passing ring (e.g. IEEE 802.5)

The token passing protocol is the oldest and most popular mechanism being used in the IBM token passing
ring and equivalent networks (4 Mbits/sec to 16 Mbits/
sec over twisted pair). The underlying principal is similar
to the big brass keys that were used on steam trains to ensure that only one train could be on a single track
line.

A special token packet (
e.g.

11111111) circulates around the ring. If a station

wants to transmit, it must wait
until it receives the token. It seizes the token by flipping the last bit and converting it to a connector
(11111110) and follows this by inserting its message

packet onto the ring

(bit stuffing is used to prevent a
token p
attern occurring in the message data).


A token being converted to a connector plus message

This transmitted bit stream then passes through each node on the ring. No other station can send since it
hasn't got the t
oken. Each station looks at the address in the message packet to see if it is addressed to it, if
not it ignores it and passes it on. If the address is that of the station it copies the packet into its buffers and
flips an acknowledgement bit at the end of

the packet to indicate receipt. The packet is repeated by each node
and eventually arrives back at the sender node which takes it off the ring, recreates the token and sends it to
the next station. Thus a station many only use the token once before passin
g it on. This 'round robin'
technique ensures fair access to the network and prevents a station 'hogging' the ring once it has got the
token. Token passing provides a reasonably fair allocation of resources to each station under high load;
packets are norm
ally kept small to prevent long messages hogging the ring. A monitor station is required to
removed damaged frames, and recreate a missing token or remove duplicate tokens due to line errors.

The only medium specified in 802.5 token passing ring
standards

is shielded twisted pair cable operating at 4
or 16 Mbps. This is the standard used by the IBM token passing ring and its equivalents. Many vendors (
e.g.

IBM) have their own high performance token passing rings for minicomputer and workstation
environment
s

(usually based on coaxial cable).




4.3.1.2

FDDI (Fiber Distributed Data Interface) Network (ANSI X3T9.5)

It has been
recognized

for some time that current bus and token passing ring networks operating at speeds of
10
-
20Mbit/sec are incapable of suppor
ting the high speed communications needs of powerful workstations
and modern applications. FDDI (Joshi 1986, Watson & Cunningham 1990) makes use of the high bandwidth
and noise immunity of fiber optic cable to operate at speeds greater than 100Mbit/sec and

over distances of
hundreds of
kilometers
.
The diagram below
shows the structure of a typical FDDI system which consists of
two independent communications rings each operating (in opposite directions) at 100Mbit/sec giving a
Local Area Networks

11

combined bandwidth of 200Mbit/s
ec. FDDI uses a Token passing protocol similar to that of the Token Ring
except that when a station waiting to transmit captures the token it transmits packets of information and then
issues a new token which the next station can capture for its messages.
Thus packets of information are
circulating around the ring followed by a token. This slight change in the protocol makes better use of the very
high network bandwidth than would be possible with the simpler (single) token passing protocol.

If the network

is broken the two rings reform to become a single ring (by 'wrapping' the connections on eithe
r
side of the fault)

thus maintaining network service. In practice both rings do not have to be taken to all stations
with less critical nodes only have one ring

connection thus saving cost.


The FDDI (Fiber Distributed Data Interface) Network


FDDI operating as a single ring when a break occurs in a link between nodes

Up to 1000 nodes may be co
nnected to an FDDI ring with a maximum spacing of 2km between nodes giving a
maximum circumference of 200km.

The major problem with

FDDI is that it is expensive
not only in terms of cabling, but the cost o
f interface cards.
It tends to be used to interco
nnect buildings on a large site with high speed networks based on 100baseT
used within the buildings.

Local Area Networks

12

5

Hubs (or wiring
centers
)

10Base2 networks using coaxial cables are usually multidropped from the cable using coaxial T connectors so
that the physical
and logical representations look similar,
e.g.
:


10BaseT connections are usually taken to a wiring centre called a
hub

which
organizes

the network into a
bus. In this case the physical layout looks like a star although th
e logical layout is still a bus,
e.g.
:


Rings often use a similar hub system with the stations connected on 'petals',
e.g.
:


This still operates as a logical ring since all traffic follows the path:

A
-

B
-

C
-

D
-

E
-

F
-

G
-

H
-

A
-

etc.

Using wire
centers

makes networks much easier to install and to identify and isolate faulty loops, e.g. a broken
cable, bad cable connector or faulty station.

5.2

Intelligent Hubs

In cases where the load on a CSMA/CD bus network is h
eavy and many collisions would occur an intelligent
(or active) hub many be used. In this case the hub has a processor and allocates RAM memory to buffer data
going to/from the stations, removing the possibility of collisions, e.g. when the destination lin
e is free the
packets are transmitted. In addition the fileserver connection may be faster (e.g. 100BaseT) than the
workstations (e.g. 10BaseT) to improve throughput. Remember when installing 100BaseT connections on a
PC to use PCI network cards (using a 3
2/64 bit bus) with DMA (direct memory access) to get the required
throughput.




Local Area Networks

13

6

Network Access: Probabilistic or Deterministic

Probabilistic
access occurs when devices compete for access, i.e. a contention bus using the CSMA/CD
protocol. There is only
a certain probability that a particular device will be granted access. There is no
guarantee of access within a specified time and under heavy network loads a device may never get access.

Deterministic
access occurs when the protocol pre
-
determines when a

device will be granted access. For
example, in a token passing protocol access is granted to each device in strict rotation (a station can use the
token once then must pass it on to the next in sequence). Thus a device will always get access (within a tim
e
which can be calculated from network speed, number of stations, etc.) no matter how busy the network is.

Effect on applications


The application area may well determine what type of network and/or protocol may be used. For example, in
real time applicat
ions (such as process control, real
-
time voice, etc.) need to have guaranteed access within
specified time limits. When attempting to use Ethernet to transmit real
-
time voice conversations one has no
idea when the data will get through (there may be long b
reaks in the conversation). In a real
-
time control
systems (e.g. car, nuclear power station, etc.) it is very important that critical messages can get through within
a specified time, e.g. a message to apply the car brakes does not get through because the
network is being
heavily used by the CD player. Multimedia applications which involve transferring large files (images, real
-
time
video, voice, etc.) over a network can also cause sever problems if there are bottlenecks either on the servers
(lack of disk
capacity or memory) or the network.

Contention busses (such as Ethernet) are fine for applications where traffic is relatively low and there are no
real
-
time applications running. Rings react evenly to heavy traffic but have the problem of maintaining phy
sical
ring integrity. The token passing bus has the advantages of both (easier to install than rings but with
deterministic access) but the software is very complex (commonly used in factory automation).



7

Internetworks: Bridges, Routers and Gateways

An

internetwork

is a collection of WANs and/or LANs that are connected together via bridges, routers and
gateways (see separate notes on
internetworks
).

A
bridge

is used to interconnect two similar LANs and acts

as an address filter, e.g. to break up a large
network into logical components. The bridge contains addressing and routing intelligence and is aware of
which station addresses are on which network. Only packets destined to stations on the other network ar
e
passed across; thus confining local traffic to each network. Splitting up a network into a number of semi
-
independent sections can assist with network management, security and the problems encountered with bus
networks using the CSMA/CD protocol.

A
gate
way

is used is used to interconnect dissimilar networks, e.g. bus LAN to ring LAN. It must contain
protocol conversion software in addition to addressing and routing intelligence. A
router

is used in where
more than two networks are to be interconnected. I
n practice a bridge may also act as a router when
interconnecting networks of the same type otherwise a sophisticated gateway may be required. The routing
information is generally provided by static routing tables which are set up by the network manager.

Con
sider the network shown below
where each office of a small company has its own Ethernet network (with
fileserver and user stations) interconnected via bridges and an Ethernet backbone. Local traffic is contained
within the subnetworks with only inter
-
of
fice traffic crossing onto the backbone. Consider if station A in the
sales office wishes to communicate with station R in the research and development office. Bridges B1, B3 and
B4 act as routers to transfer the information (note that it is important that

there is only one path between
stations!
).

Local Area Networks

14


Ethernet backbone with bridges to subnetworks

If an internetwork is configured correctly the majority of traffic on the network is local to the semi
-
independent
secti
ons with the bridges and gateways handling a (relatively) small amount of long distance traffic,
e.g.

data
transmitted to a remote minicomputer or mainframe for analysis. In extreme cases (
e.g.

when security is
paramount) the network sections can be physic
ally independent and linked via a bridge only when necessary
(using a physical switch). In practice the bridges, gateways and routers may be 'off
-
the
-
shelf' devices or a
fileserver equipped with two or more network cards and suitable software (
e.g.

Novell
NetWare).

Local Area Networks

15

8

Distributed Environments

A modern computer configuration tends to consist of a number of individual computer systems connected via
a network to form a distributed environment:

(a)

The network: communications hardware and software which forms the

backbone of the distributed
environment.

(b)

User workstations: terminals, PCs and/or professional workstations running highly interactive software
tools, e.g. word processing, spreadsheets, CAD design, CASE tools, etc.

(c)

Servers: powerful (relative to the us
er workstations) computer systems which provide general services
to the network users,
e.g.
:

(i)

fileservers: hold general operating system files, compilers, user files, etc.

(ii)

print servers: provide a printing service possibly using very high quality printers
.

(iii)

database servers holding
centralized

databases.

(i v)

computational server: some commercial, industrial or scientific applications require a
powerful
centralized

to provide computational power beyond the capacity of the user
workstations, e.g. heavy floating

point numeric processing, vector processors, etc.

(v)

X terminal server: X terminals provide the X windows environment of a modern
workstation to programs running on a remote host server (they have no computing power
of their own). X windows applications mak
e a severe demand on the host in terms of
processing power and main memory requirements (typically 8Mbytes/X terminal).

In practice a server may be a high powered PC, a
specialized

workstation (without a user), a
minicomputer, a mainframe or supercomputer
. Sometimes a server may perform several
roles, e.g. a file server, printer server and a database server (care must be taken not to
overload servers or bottlenecks will form).

(d)

Bridges and gateways: in a distributed environment care must be taken not to ove
rload the network
and to provide adequate security for sensitive information. Splitting a large network up into a number
of semi independent networks linked by bridges and gateways can assist

with these problems.

8
.1 Performance factors in a distributed en
vironment

Distributed environments can be very complex and the overall performance depends upon many factors.

Performance of the network or networks


This is dependent upon the physical network configuration and speed, the communications protocol used,
nu
mber of bridges and gateways, etc. A major problem is that the majority of the local area bus and ring
networks installed today are based on technology over ten years old,
e.g.

Ethernet was developed in the
1970's by Xerox. The communications speed (typica
lly 10Mbit/sec) is incapable of supporting the high speed
communications needs of powerful workstations and modern applications, e.g. where PCs are rated at 10 to
20 Mips (millions of instructions per second) and professional workstations at 50 to 200 Mips
. Moving to
modern high performance networks such as FDDI is clearly a requirement but often impossible due to cost.

Network configuration


The number of user workstations, their distribution over the network(s) together with servers, bridges,
gateways, e
tc. Modern network have hardware and software tools (which can keep track of network traffic,
files used, etc.) to assist with network configuration.

User workstations


Computational power, main memory size and disk size:

(i)

computational power: in general
only effects the performance as seen by the individual user; but
high powered workstations may be slowed down by bottlenecks elsewhere, e.g. by a slow
networks or overloaded servers.

Local Area Networks

16

(ii)

used software is held on local disks (needs careful management of new so
ftware releases). A
cheaper alternative is too have a disk of sufficient size to hold the operating system swap space
(otherwise multi
-
tasking and virtual memory page swapping takes place over the network, see
diskless nodes below).

Fileservers


The number

of fileservers and their power in terms of processor performance, main memory size and disk I/O
performance:

(i)

computational power: power sufficient to cope with network and disk I/O.

(ii)

I/O bandwidth: a good as possible to support disk and network I/O.

(iii)

siz
e of main memory: very important
-

sufficient to buffer information.

(i v)

disk speed and size: speed is very important (can be improved by using caching techniques) and
size sufficient to hold files, print queues, etc.

(v)

network interface: good, possibly with D
MA (Direct Memory Access) facilities.

The distribution of software packages and user files around the fileservers is critical:

(a)

complex intensive
centralized

tasks could well require a dedicated fileserver, e.g. an advanced
database environment or the anal
ysis of large engineering structures using finite element mesh
techniques;

(b)

spreading the end
-
user files around the fileservers prevents overloading of particular fileservers (and
if a fileserver breaks down some users can still do their work).

Common prob
lems


Clearly great care is needed in configuring a distributed environment with a slight error giving the impression
of 'clockwork' powered machines. Common problems (often due to lack of funds) are:

1.

Network speed too slow for modern workstations togethe
r with poor distribution of files, i.e. workstations
having to get common files from servers, e.g. PCs getting windows from a server rather than their own
disk and UNIX workstations paging over the network.

2.

too few fileservers for the number of user workst
ations and/or poor distribution of fileservers across the
network;

3.

too little main memory on fileservers causing bottlenecks in the accessing of
centralized

file systems;




9

What network to buy?

There is no easy answer as to which type of network is most

suitable for a given application. Networks based
on baseband bus (e.g. Ethernet) were the earliest and are still widely available and used. The mechanism is
robust and uses many well proven components, e.g. from cable TV

and telephone technology
; opponent
s
would argue that the CSMA/CD contention protocols used are unsuitable for high traffic loads, cannot support
voice traffic and severely limit the maximum length of the bus.

However, intelligent hubs can overcome some
of these problems.

Rings are capable

of covering greater distances and of reacting more evenly to heavy
traffic but are typically more expensive and have suffered reliability problems in the past (a single poor
connection on the ring can cause havoc and be very difficult to track down). Star
s have the great advantage
that they can often be installed using existing cabling but HAVE the disadvantage of all
centralized

networks
-

dependence on a master station

or hub
. Broadband networks can handle enormous traffic flows at very high
speeds; they

are the only effective mechanism at present which can mix data, voice and image traffic; but they
are extremely expensive.