25

3 Local area networks

A local area network (LAN) is a number of computers and computer
peripherals (disc storage devices, printers) connected by high speed data
lines within a building or adjacent buildings.

LANs originated as ways of interconnecting diverse electronic equip-

ment within an organisation, which would permit local processing while
also giving access to other devices connected to the network (inter-station
message transmission without the use of a central host computer, access
to various file stores). Potentially, LAN traffic is not restricted to
character transmission: some architectures would permit the transmis-
sion of voice and video signals. The total distance spanned by a LAN can
be up to a few kilometres.

Although LANs are conceptually very different from the conventional

"star" data processing network of terminals connected to a powerful
central host computer, in practice there is often still very little perceived
difference to users or even to applications designers.

Many LANs, and certainly those of the type which were used for this
project, are intended mainly for office automation applications such as
word processing, small-scale management information systems and
"electronic mail". Their user stations are normally fairly conventional
microcomputers of the personal computer type. Typically, their
applications would involve multi-access to a large number of small files.
Nestar, the supplier of the networks used for the current project, were
amazed when they discovered that we intended to store, process and
search files of ten, twenty or more megabytes.

Some of the terminology which is used may be unfamiliar: computers

used as terminals on a network are often referred to as user stations;
computers which control disc storage are called file servers, and those
which control hard copy output are print servers. Networks may be
interconnected by means of gateways.

There is a large and growing literature on the subject of LANs. Cheong

and Hirschheim [1] gave a not too technical introduction to the subject as
26 3. Local area networks

it stood in 1982. There is a more recent brief survey of the field in [2]. Mel
Collier [3] has given an introduction with an emphasis on potential
library uses.

A brief summary follows of those aspects of LAN design and architecture

which have a bearing on the present project.

3.1 Types of LAN

LANs may be compared with respect to the method of interconnection of

the participating devices (topology and connection medium) and to their
methods of data transmission and control.

Topologically, LANs may best be classified as being either rings or trees;

{star and bus topology are often distinguished, but both of these may be
considered as trees). In a ring network devices are connected like beads on
a necklace, the most significant examples being LANs of the Cambridge
ring type. The distinguishing property of tree networks is that there are no
closed loops.

Independently of the topology, the connection medium may be coaxial

cable (as used for television signal transmission), twisted pairs of
conductors as used for telephone signal transmission, ribbon cable of
perhaps ten conductors, and fibre optics (where light is used as the carrier
instead of electrical signals). Most currently available LANs use either
coaxial cable or twisted pair. Coaxial cable allows higher data rates, but is
somewhat more expensive and less robust than twisted pair.

Data transmission is generally in "packets" of a few hundred to a few

thousand bits. Each packet contains source and destination addresses as
well as "rear' data. Some LANs use packets of fixed size: the size chosen
has to be a compromise between wasting capacity by splitting long
messages into multiple packets, and wasting it by having to transmit more
than is required when a message is very short. The speed of data
transmission is between a few hundred thousand and several million bits
per second. It should be noted that the total amount of data which can be
transmitted over a network is considerably less than that given by the
"raw" data transmission rate, because of the necessity for transmitting
control information and for avoiding collisions. A network with a raw
data speed of five megabits per second might be able to achieve a
maximum effective overall capacity of two to three megabits per second.
3. Local area networks 27

However, transmission capacity is rarely a limiting factor. Transmission

accuracy is generally high, particularly in ring networks; in any case,
error detection codes are incorporated in the packets, so that retransmis-
sion can take place automatically in the event of an error.

As far as control is concerned, a general philosophy has been to avoid

centralisation. A good LAN installation should continue to work if a
number of components fail or are disconnected: in some systems this is
almost an inherent property; in others it can be achieved by, for example,
the duplication of devices such as cables and file servers. In ring systems
the failure of a device is likely to cause the entire network to fail unless
such precautions have been taken.

There are two widely used methods of controlling access by individual

stations to the network.

One, used in the Cambridge ring, is to circulate a "token" or "slots"

round the ring: a station can transmit when, and only when, it holds the
token. This is simple and efficient, and should not be thought of as greatly
decreasing the capacity of the network, because the token circulates
extremely fast.

The other method is used by LANs of the Ethernet type. {Ethernet is a

public specification which resulted from preliminary work by the Xerox
Corporation; details were published jointly by Xerox, Digital Equipment
and Intel [4].) With Ethernet-type networks, access is controlled by a
method known as Carrier Sense Multiple Access with Collision Detection
(CSMA/CD). It is analogous to confrontation avoidance in felines
[5, p219], and works roughly like this: a station wishing to transmit
listens to the network; if it detects no other signal, it transmits. It then
reads its own data back to check that no collision with another packet has
occurred. If there has been a collision, it emits a "jamming" signal, waits
for a random period (up to a few milliseconds), and then attempts to
retransmit. The process is repeated until a successful transmission has
been achieved. Clearly CSMA/CD is far less simple than token passing,
but it works very satisfactorily. It imposes somewhat more of a limitation
on network size than does token passing, but this can be minimised by
using a fairly large packet size.
28 3. Local area networks

3.2 LAN or multi-user minicomputer?

In choosing between a LAN and a multi-user minicomputer there are

two factors to be taken into account: cost and performance.

3.2.1 Relative cost

It had been hoped to give here some comparative hardware and software
prices to help prospective purchasers in their choice of configurations for
the automation of small libraries. For a combination of reasons it is not
possible to do this realistically.

HARDWARE

Although suppliers of LANs and of minicomputer installations will often

quote "list prices", these may bear little relation to what a corporate
customer would have to pay. For example, a customer who can qualify as
an original equipment manufacturer (OEM), may obtain equipment at
perhaps half the list price; some suppliers would also give "educational
discounts". It is likely that at the present time there is little difference
between the cost of LAN (or other multiprocessor) and single-processor
hardware to carry out data storage and retrieval on the scale needed by
small to medium libraries. The current list price for a LAN installation of
the type and size on which the Okapi prototype operates is given in
Section 3.3.1.

SOFTWARE

Here also it is not possible to make a comparison, but the reason is that,
with one exception, there is no commercially available library automation
software which is designed to operate on a LAN. (The exception is the
large Geac installation at New York University [6].) There is a certain
amount of library software for stand-alone microcomputers, and some of
this might be adaptable for LAN configurations.

3.2.2 Performance considerations: LAN advantages

The potential advantages of LANs stem mainly from the fact that each
user has a station which is very much like a dedicated desk-top computer.
Multi-processor systems — a number of processors in one box — are very
similar, but may not provide the same high data transmission speed to
user terminals.
3. Local area networks 29

The implication of the single-user-single-processor (e.g. LAN) confi-

guration is that highly interactive programs can operate. When the
processing is time-shared by a single processor, as in a typical
minicomputer environment, it is not possible to operate programs which
involve instant responses and finely tuned dialogue, such as computer
games with elaborate interactive graphics, or even screen-based word
processing programs.

This is well illustrated by the process of searching conventional online

reference retrieval systems such as DIALOG or Datastar where, apart
from problems with noise, and losing telephone contact, the system's
response time is highly variable and often very slow, and data
transmission speed rarely attains a hundred characters per second: they
cannot engage in the kind of rapid exchanges which occur in
communication between humans, or between a human and a computer
game on a micro. Interaction has to be by means of a well defined
command language, as any other approach requires a much higher degree
of responsiveness.

For an online catalogue, the local processing power of a LAN allows

instant response to some user needs — help messages or prompts can be
stored in local memory and displayed instantly when needed. The
importance of this lies in the fact that programs can be made to respond
much more finely to specific user needs and to the current state of a
search: for example, when the user is entering a specific item search in
Okapi the message which gives possible next actions changes instantly
according to whether the user is entering a title or an author and
according to the amount of text entered (Section 7.5).

It must be emphasised that such highly interactive programs are difficult

to design and to write (Section 9.4.4).

It is of course true that an IR system on a LAN will still have search

speeds which vary according to the load on the system, because of varying
loads on the network and, particularly, on the file server, but much of the
human-machine dialogue takes place without any access being made to
the network or disc store. User input can be validated for acceptable
structure, instructions or help can be given rapidly and automatically. To
some extent, search programs can be made so that network access takes
30 3. Local area networks

place when the user is occupied reading what is on the screen or keying
something in.

3.2.3 Performance considerations: shared processor advantages

From the point of view of software development and file maintenance

there are certain advantages in the multi-user mini if it is not too heavily
loaded. Both the processor and disc access are likely to be very much faster
than those available with microcomputer based LANs. The mini is
unquestionably superior for batch processes such as source file and index
maintenance, and in some respects also for program development.

In the Okapi system file inversion (creation of indexes) for a file of 90,000
bibliographic records takes many hours, although the inversion programs
are written so that certain processes can be done concurrently using
several stations simultaneously (Section 5.7.6). In program development
we became inured to quite small compiling and linking jobs taking many
minutes.

The mini's superiority for software development stems also from the fact
that most minicomputer suppliers will be able to provide well-proven and
sophisticated system software such as compilers, operating systems and
file management systems, together with libraries of high level routines,
sorting procedures for example, which can be included in applications
programs. Micro-based LANs intended for office automation are not
supplied with such system software, and it was not possible for the team
to make much use of the software which was provided. To obtain
satisfactory performance from the Okapi programs it was necessary to
work in assembly language, and to write an extensive library of
procedures for input and output, comparison, sorting and file
management.

3.3 The project LANs

3.3.1 General description and cost

The LAN system originally chosen for this project was a Nestar
(formerly Zynar) c 'Cluster/One". At an early stage in the project this was
replaced by a "PLAN 4000" from the same supplier. Each of the two
PLAN 4000 installations which were used consists of a file server, a
number of user stations, a Winchester disc drive of 137 megabytes, a
3. Local area networks 31

printer, and also a 40 megabyte cartridge tape drive for archiving the
contents of the Winchester.

The file server is a microcomputer with a 68000 CPU, provided with a

number of system programs enabling applications programs, which
reside in the user stations, to use the network to read from and write to the
central Winchester disc store.

The user stations are Apple He microcomputers. The PLAN system also
allows IBM PCs to be used as stations, but the Apples were very much
cheaper and have a keyboard which is more suitable for people unfamiliar
with computers (almost the same layout and number of keys as an electric
typewriter). On the other hand, Apples are extremely slow and primitive
compared with more recent microcomputer designs. As it turned out, this
has scarcely been a constraint. The Apples were supplied equipped with
Z80 processors enabling them to run much of the software written for the
widely used CP/M operating system. A version of CP/M is supplied by
Nestar, as well as other operating systems for standard Apples and for
IBM PCs.

Stations can be connected to the network in a fairly arbitrary way,

provided that there are no closed loops. Data transmission is by coaxial
cable. Line isolation devices (LIDs) are used for connecting stations to the
network. These are of two types, active and passive. Each active L I D has
six ports, so it can be used either to connect five additional stations to the
network or to connect four stations and also provide a cable to another
part of the building; thus an economical configuration might resemble a
sequence of stars emanating from LIDs on a bus. (The passive LIDs have
four ports, and do not need a power supply, but are of limited use since
they can only be used as the final node in a branch of the tree.) A station
can be connected or disconnected, or can fail, without affecting other
stations.

A typical layout is indicated in Fig. 3.1.

Cost

Based on January 1985 list prices from Nestar, and average figures for the
cost of the Apple He user stations, the cost of the hardware for an
installation of the type on which Okapi was tested breaks down as follows:
32 3. Local area networks

file server, with 137 megabyte disc drive £18500

(a 60 megabyte disc drive costs £3000 less)
10 network interface cards 4950
10 Apple He's, with monitors, 80-column
cards and Z-80 cards 7000
4 active LIDs 1980
2 passive LIDs 100
1000 feet of coaxial cable and connectors 150
printer 500

total £33180

The file server price includes file server system software, but the cost of
operating systems, compilers etc, for user stations is not included. To this
would have to be added the cost of installation, and about 15% per annum
for maintenance.

3.3.2 Performance

Early in 1983 some tests were done on the Zynar Cluster/One network
which involved a number of stations making random reads from a fairly
large file. It was found that over the whole network, almost regardless of
the number of stations (one to four), only about six kilobytes per second
were being transmitted. The raw data transmission speed was almost
certainly irrelevant, but it was difficult to determine whether the file
server software or the physical speed of disc access was to blame. (It
turned out later that the limiting factor had been using the operating
system CP/M to control network access.) This low data transmission rate
would have been an important factor in the design of the IR system
(Section 1.3).

The PLAN 4000 is somewhat faster than the Cluster/One, but the file
server software was at first a limiting factor. Whenever several stations
were carrying out processes which made intensive disc access via the
network, from time to time there would be a pause of a second or so while
the whole system seemed to be sorting itself out. By the time the second
PLAN 4000 was obtained there was a later release of the file server
software in which network and disc access seem quite nicely balanced.
Nestar later informed us that there had been a bug in the network
queuing procedure. It now became possible to feel more confident about
3. Local area networks 33

including functions, like Boolean OR, which give rise to intensive disc
activity.

Nestar supplied routines for disc access via the network which by-passed
the CP/M operating system. These were incorporated in the search
program when the prototype was nearly ready and gave a further very
considerable improvement in the speed of operations involving disc
access.

file server

Figure 3.1 LAN diagram

VI 3. Local area networks

3.3.3 Reliability
Things which can fail are the disc drive, the file server, user stations and
connecting cable and LIDs. All hardware and system software was
covered by a maintenance agreement at 15% of purchase price per year,
and this gave service usually within about half a day.

The Cluster/One disc drive had timing slots which were exposed to the
atmosphere and gradually filled with dust, and these had to be cleaned
every few months.

After running almost continuously for more than a year the disc drive of
one of the PLAN 4000s failed. Zynar replaced the entire disc unit and all
files were restored from backup tapes, losing almost nothing. The user
stations have been surprisingly reliable — with the ten Apple He's used
with the PLAN 4000s there has been one keyboard failure and one failure
of a Z80 card. The power supply to one of the LIDs failed. Failure of a
cable will only result from physical damage, and the coaxial cable is
robust unless bent vigorously.

There have been failures of the mains supply. There is no automatic

detection of incipient power failure; when power is restored the disc drive
and file server start operating again, but some new data may have been
lost. For our purposes it was certainly not considered necessary to have an
emergency power supply. A normal, fairly "dirty", mains supply has
been used for one of the installations, although the second one was
connected directly to the incoming cable on a separate circuit. There has
been no problem with static electricity, and one of the networks has
functioned in ambient temperatures between about 5 and 35 degrees
Celsius.

3.3.4 Some technical information on the PLAN 4000 LAN

The PLAN 4000 has layers of network protocols following the

International Standards Organization (ISO) guidelines for Open System
Interconnection (OSI). The physical layers use a token passing protocol
(Arcnet), and the intermediate layers are like Ethernet. The network has
a "distributed star" topology, but the stations behave as if they were
connected in a logical ring. The raw data transmission speed is 2.5
megabits per second, and the maximum effective network capacity is
about 1.6 megabits per second (roughly 200,000 characters per second).
Packet size is variable up to 256 bytes.
3. Local area networks 35

The network capacity considerably exceeds the amount which a file

server can read and transmit when ''random" disc accesses are being
made. The 137 megabyte disc drive has a mean access time of 42
milliseconds, and a "cache" or buffer of four kilobytes. Up to four such
disc drives can be controlled by a single file server station. It is possible to
use more than one file server on a network. Direct communication
between user stations is also possible, as is the use of local floppy or hard
disc drives at individual stations.

Stations are connected to the network by means of a network interface

card. This has a two kilobyte I/O buffer, and contains PROMs which
control network access by the station. The theoretical maximum number
of stations on a single network is 255, but for most practical applications it
would not be advisable to exceed about twenty to thirty stations for each
file server [7,8].

3.4 Data security

This section applies to any computer installation. We have thought it

worth including, because security is normally the responsibility of a
computer operations manager, and as installations are becoming cheaper
many are being operated without a professional manager.

It is of course vitally important to ensure that there are always backup

tapes of all important data. With mainframe computer installations there
is normally some archiving done every night and a complete dump every
few weeks, organised in such a way that in the event of a complete disc
failure not more than a day's work will be lost. With a small disc drive the
simplest and most reliable procedure is to take two complete backups
daily (if some file has been altered on that day). This is rarely practicable
with discs of forty megabytes or more. One recommended method is to
make a weekly dump of the entire disc and a daily backup of every file
(pathname) that has been written to since the last complete dump. Nestar
provide a program which enables selective archiving, and the system
timestamps pathnames with the first time they are written to on a given
day.

All the networks which the project has used (Cluster/One and PLAN
4000s) are equipped with their own built-in tape drives. The cartridge
tapes (similar to audio cassette tapes but somewhat larger and able to run
at a very high speed) are expensive, but fast and reliable. It usually takes
36 3. Local area networks

about ten minutes to write thirty or forty megabytes to a tape: the exact
time depends on the number of pathnames. The archiving program
provides a verification option, in which the tape is re-read and its
contents compared with the contents of those parts of the disc of which it
is a copy. Verification is much slower than writing since a rather small
buffer is used and the tape has to keep stopping and restarting. Whenever
possible, tapes were verified, and there have been several "verify errors".
It is not possible to do anything with the tape drive while the file server is
running, so backup has to be carried out when the network is not being
used.(1)

Apart from magnetic tape, other media which might be offered for
backup by suppliers of microcomputer installations include videotape
and floppy discs. The former is satisfactory provided that the amount of
stored data is not too large: the drives are cheaper than those for magnetic
tape cartridges, but it is not possible to do selective saving and restoring.
Floppy discs are still only suitable for the smallest installations, if only
because the capacity of a disc is unlikely to be more than a megabyte or so.

3.5 The design of applications software for LANs

A major advantage of LAN distributed processing has been mentioned

above: it is possible to use programs which are highly interactive in the
sense that they involve consistently fast responses to user input. Thus,
software designed for single user computers, like word processing
systems and "spreadsheet" programs will usually work almost as well on
a LAN as on a dedicated microcomputer. Except when disc access is being
made, the stations are single user computers.

However, library applications, except on a very small scale (when it is

doubtful whether computers are of any practical use anyway), need
access to comparatively large files. Then, it is likely that, just as with
multi-user-single-processor systems, processing may become "disc
bound". That is, the time required to read from or write to the central
disc store may often be more important than the speed of the processors
and the data transmission capacity of the network. Thus the design of the
disc access hardware and software will be a very important factor, and
this is something over which the purchaser has no control.
(1) By using a second file server on the PLAN network a high degree of automatic security
can be provided. This would be important if a LAN were to be used for, say, circulation
control.
3. Local area networks 37

This project entailed building an IR system in which up to twenty or so

users could simultaneously retrieve information from a file or files of up
to several hundred thousand bibliographic records. Such a system is
closely analogous to online and in-house reference retrieval systems,
almost all of which use multi-access to single mainframe or minicom-
puters. There has been some disagreement over whether these IR systems
tend to become disc bound or processor bound, but undoubtedly disc
access is an important factor in affecting response times. Although data
transmission over LANs is very much faster than the transmission
between a mainframe and a terminal, regarded as devices for allowing
terminals to access disc files, microcomputer based LANs are consider-
ably slower than many mainframe installations. Consequently the project
software had to be designed so that disc access was reduced to a minimum.

T H E LAN AS A PROGRAM DEVELOPMENT ENVIRONMENT

As mentioned in Section 3.2.3, system software for microcomputers is

rudimentary in comparison to that available for mini and mainframe
installations. In particular, there are few, if any, high level programming
language systems which can be used to write programs for applications
which call for high processing speed and fine control over memory
resources and timing. Had we been working with a minicomputer system
we should undoubtedly have used an implementation of one of the widely
available high level languages which are suitable for applications
involving character processing, such as C or PL/I (or possibly Pascal,
which is easier to read but somewhat restrictive). This would have had
the advantages of saving development time, improving portability and
reducing the effort of documentation. There are, of course, high level
language systems for micros, but we did not consider any of them to be
suitable unless used in combination with a considerable amount of
assembly language. (One of the Okapi programs — the second stage of the
file inversion process (Section 5.7.6) — is written in a mixture of
compiled MicroSoft BASIC and Z80 assembler, but this is a batch
operation and a very simple process which can afford to be fairly
inefficient.) All the remaining programs have been written entirely in
Z80 assembly language. No doubt this unsatisfactory situation will
change in the fairly near future.

On the positive side, the highly interactive facilities made possible by the
micro environment meant that the team could use a word processing
system, WordStar, for almost all keyboarding of program source code
38 3. Local area networks

and other textual data. This is very much easier than using the line editors
which are usually provided in time sharing systems.

References

1 Cheong V E and Hirschheim R A. Local area networks: issues,

products, and developments. Wiley, 1983.
2 Local area networks — how computers talk to each other. Institution of
Electrical Engineers. Digest number 1984/50. 1984.
3 Collier M. Local area networks. British Library. Library and
Information Research Report 19. 1984.
4 Foot B. Ethernet — basic principles. Institution of Electrical Engineers.
(Undated).
5 Leyhausen P. Cat behaviour. Chapter 19: Territorial behaviour and
rank. Garland Press, 1979.
6 Persky G et al. A Geac local area network for the Bobst Library.
Library HI TECH (6), 1984, p37-45.
7 [Nestar Systems]. Unpublished communications from staff of
Nestar Systems, 1983-1985.
8 Frith W R. A layered approach to local area networks. New
Electronics 17 (24), 1984, p64-66.