Magnum One/Magnum One (Mid-American Digital) (Disc Manufacturing).iso/d8/networks.arc/NETWORKS.TXT

Text File | 1988-02-06 | 23KB | 519 lines

How to Evaluate Networks
by Ken Goosens

Copyright (c) 1985, Capital PC User Group Inc.
This material may be republished only for internal use
by other not-for-profit user groups.

Published in the May 1985 issue of the Capital PC Monitor.


The purpose of this article is to educate users on how to evaluate networks,
or "What you should have asked before you bought."  This article provides
the basis for comparing networks but does not review any particular network.
A check list of particular items to keep in mind when shopping for networks
follows this article.


Why Network?

Networking PCs together allows resources to be shared, such as printers
and hard disks.  Using a computer no longer ties up all the resources
on it.  Fewer peripherals need to be purchased since unused resources
on any machine can be used by others.  Disk storage space can be conserved,
since only one copy of the software need be kept on the network.  Applications
stored on the same hard disk can be run at the same time instead of having
to wait until the PC with the hard disk becomes free.  Users can work
on a single master copy of a data base rather than work on separate copies
that have to be consolidated later.


Why Not Network?

Networks make applications inter-dependant and increase the need for
support, cooperation, and management.  Mistakes and problems now affect
many users rather than a single one.  No networking standard exists in
the marketplace.  Networks are relatively immature products that still
have significant limitations and problems.


Understand Before You Evaluate

Don't start evaluating the detailed implementation until you first understand
what makes a network better designed.  A bad implementation of a good
design can make a network unworkable, but the best implementation cannot
compensate for flaws in the design.  Compare designs, then performance.


What Can be Shared?

The essential purpose of a network is to connect computers so that they
can share resources.  The number one item on everybody's list of needs
is

o    Mass storage, chiefly hard disks.

There are two features of disk sharing that are highly desirable.  First,
you want an open architecture in which virtually any hard disk which
will run on a stand alone machine will be sharable on the network.  Ask
whether the hard disks you already have are shareable.  Proprietary hard
disks lock you into a limited and expensive line of products and do notlet you take advantages of improved products.  Also, you want to be able
to share all the drives on a hard disk.  Some networks limit you to a
single shareable drive on a hard disk.

You basically want to be able to share whatever resources that users
have to wait for because they are attached to a machine in use.  After
hard disks, people most want to share

o    Printers, both high speed and letter quality.

Most networks will share a printer.  But some require special software
and a personal computer to be dedicated solely to printing.  Some will
not share a letter quality printer; others will not work with a sheet
feeder.  Some will not share a printer attached to a serial rather than
a parallel port.  Some charge you $700 for every printer station on a
network, others charge you nothing extra.  All printing systems must
have some way to queue print jobs from different stations on the network,
so that simultaneous printer requests do not get their output mixed together.

Other possibilities to share include

o     A connection to a mainframe.

If you have IBM mainframes, for example, you can cable one computer directly
to you mainframe which then will act as a "cluster controller" for other
stations on the network.

o    Modem.

Few people at any one time need a modem, so can single shareable modem
on the network may be enough for 10 people.

o    Plotters.

Most plotters sit idle most of the time, so a shared with say 8 pens
and an automatic paper feed can be very useful on a network.

o    Electronic drives.

You can install a huge electronic disk on a server which any station
can then use as a fast scratch work space, say 2 megabytes on an AT.

o    Tape drives.

A 9 track tape drive is an invaluable resource for transferring data
from larger computers to PCs, and a shared one means people do not have
to wait for the machine with the tape drive attached to become free.

o    The computer itself.

A computer that is sitting idle on a network can be commanded remotely
to run jobs from other stations on some networks.

Understanding Network Servers

How are networks designed so that devices are shareable?  Every shared
device has to be attached to a network "server" which will decide who
gets use of the device and will process requests for its use.  A server
or "super station" on the network makes its resources available to computers
on the network.  What a user can sit down and work on is called a work
station.  Most personal computers are run "stand alone", so that they
have no need for building in networking.  To upgrade a computer to belong
to a network requires the addition of a network card.

There are three different types of servers that can be used on a network.
Some networks have a special class of servers which are specialized and
powerful computers.  Devices are shared by being attached to this specialized
computer, so that super and work stations are physically different types
of computers.  For example, the super station may be an ALTOS 80286 running
XENIX whereas the work stations are IBM PCs running DOS.  Other networks
allow any work station to become a super station by declaring its resources
to be public, so that the distinction between a work and super station
is logically determined by the role it plays in the network.  (A super
station may, however, require a different network board in it.)  Finally,
the server may be just an electronic "box" that is attached to the network
independently from the computers people use.

By far the most popular design for network servers is to upgrade work
stations to super stations.  The main reasons for this popularity are
that users can then convert from stand alone PCs to a network without
having to purchase different types of equipment and that they can easily
share the resources already on the stand alone machines.  The most common
rationale for trying to network is that long running jobs tie up all
other applications on a hard disk.  Simply letting more than one machine
use the same hard disk at the same time solves this problem.  People
already have PCs and resources they want to share, and requiring the
purchase of another specialized computer and new peripherals just to
share these resources makes the initial costs of networking very high:
say $12,000 to network 3 PCs rather than $2000.

Making a work station into a super station is economical, but can have
disadvantages for performance and reliability.  The unreliability is
that people are used to just sitting down at a PC and doing whatever
they want.  Rebooting a super station breaks the network connection and
disrupts every work station using its resources.   Running jobs on the
super station can slow down the performance of the network for all users.
The overhead of running a network has to consume computer resources somewhere,
and non-network jobs will compete for its processing time if a super
station is not dedicated to this task.  Any network that have more than
a few work stations and that must have resources available will have
to have a server dedicated to the network.  A black box server is the
most reliable because you just turn it on and it is available no matter
what is done to any computer on the network.  But boxes like this usually
work with only one particular line of equipment.  Specialized computers
usually have much more powerful processors and faster hard disks that
are designed to perform adequately when shared by multiple users.  A
stand alone machine will have to have excess capacity, power, and speedin order to service adequately multiple users.  A work station that is
merely adequate for a single user will be inadequate as a super station.


A Good Design for Sharing

A good network has to be able to

o    share the resources that people on different work stations need
     simultaneous access to, and

o    perform well under heavy demand by multiple users.

Who needs concurrent access to what from where?  That is the primary
question to answer for any network.  Every phrase generates requirements.


First, from where.  What are the types of work stations should people
be able to use?  Do you need 8 bit computers running CPM or Apple computers,
let alone the different IBM computers, including the Junior, PC, XT,
AT, 3270, and 370.  Any answer other than just the PC and XT will have
problems.  Only in 1985 is the AT being well integrated into PC networks.
Very few networks let you mix IBM and non-IBM PCs and only a few support
the 3270 or 370 PC.

Second, to what.  For most users the only mandatory sharing is the same
hard disk.  But will any hard disk do, or do you want to share the particular
hard disk you already have?  Networks range in what hard disks they can
share from a single, proprietary hard disk sold with the network, to
virtually any hard disk that will run on work station.  Some networks
will share only one drive on a hard disk.  Some will share only an entire
drive, whereas others allow a subdirectory alone to be shared.  The moral
in evaluating a network is to

o    ask whether you can share your particular device on a particular
     machine to do specifically what you want.

Don't just ask vaguely whether a device is sharable somewhere, somehow
on the network.

Realize that performance is inherently degraded on a network compared
to stand alone, because of network overhead and because of contention
for the same resource by multiple users.  You might as well bite the
bullet:

o    make your server stations top of the line performers.

Get high capacity hard disks with cache buffers and fast access times
and use computers with more powerful processors.  However, I don't recommend
specialized computers.  These tend to be very expensive and much more
difficult to get serviced, have few people developing improved equipment
for them, and lock you into a particular product line.  Instead, use
higher performance equipment that gets broad market support, like IBM's
AT computer, equiped with a third party hard disk.  You are much lessvulnerable and can easily replace a server that fails.


Multi-User Protection

A computer system that gives multiple users simultaneous access to the
same resource must be enhanced to solve new types of problems created
by multi-users.  The simple fact is that programs that run perfectly
well on single-user systems can turn into disasters.  Three all too real
examples.  A program uses work files.  When two users run the same software,
the work files of one user overwrite the work files of the other user.
Or, a program reads in the file allocation table (FAT) telling what files
are stored where, figures out where unused space is, and writes out the
revised table only after the application is done.  Two users who enter
the same application at the same time read in the same FAT, and then
use the same space for two different new files since each thinks it is
free, thereby intertwining their files and ruining both applications.
Finally, two persons both load the same file and revise it.  The last
one to write is the copy stored and the first person's changes are lost.
Every one of these types of problems will occur on a network unless the
network is enhanced over and above DOS.

o    A multi-user system without adequate and new multi-user protection
     will be a disaster far worse than not networking.

Manufacturers of software will say that their software was never designed
to work on networks and network manufacturers will say limply that some
software will not work properly on their network.  But the bitter truth
is that users will have to discover most of these problems for themselves.

A good network will recognize multi-user problems and include features
in their design that solve them.  The main solutions include

o    Several users can write to the same drive at the same time without
     getting their files cross linked.

Otherwise, very radical limitations will have to be instituted to ensure
that no more than one user is every writing to a drive at any time.

o    Passive record locking is supported.   Without any change to the
     application software, over users are prevented from using a file
     while one user is updating it.

Users who can revise files will simply overwrite each other's changes
if there is no lockout.  By "passive" I mean that neither the user nor
the software need do anything actively to lock files: the network handles
the locking and unlocking automatically.

Realize that many networks that claim to support features like file and
record locking in fact only provide the "hooks" so that only software
that is especially written for the network - both to set and respect
locks - has locking capabilities.  Ordinary single user software will
typically have no locking and nothing on the network will prevent badly
behaved software from simply ignoring locks placed by other software.
o    New files created by different users are stored in different subdirect-
     ories.

This is an ingenious design to overcome the problem that single user
software will create work files with the same name.  By storing each
of these files in a different subdirectory and keeping track of the sub-
directory for each user (e.g. the subdirectory is the user's name), everyone
running the software will have a different copy of the work files and
so will not destroy each others work.


Multi-User Security

A computer used by only one person needs little security.  But having
many people can potentially use the same resources makes it much more
important to be able to control who has access and what they can do.
In particular, you must be able to distinguish
o    Unauthorized users who should be able to do nothing.

This is usually handled by having a procedure that requires a name and
password for logging onto the network and for getting access to resources.

o    Users who can read files but not revise them.

o    Users who have the authority to create and revise files.

A good network allows differentiation of both users and groups of files.
Yet some networks do not differentiate users.  Then shared drives are
either read-only for all users and read/write for all.  This makes it
impossible to establish levels of authority on the network.  Some networks
allow you to specify what drives a user can access, but an even better
design lets you specify access by groups of files.  For example, you
might way to put all data files used by a complex application on the
same drive, let each of three people key data be responsible for keying
data into different files, let one person only run reports, and have
a supervisor who can do anything to all the files.


Networked Versions of Software

There are two approaches to running software on networks.  First, the
network can include special designs so that single user software will
run properly on the network.  Or, software can be especially written
or modified to run on networks.  The marketplace reality is that most
applications that most users want to run were designed only for stand
alone.  Few software developers will undertake the expense of developing
special networking software until a network captures a significant portion
of the market.

There are several types of network applications that users should look
for.

o    Communications - terminal to terminal, and electronic mail.
You might want to broadcast messages across the network, either at boot
up time, or even to interupt processing without destroying it.  You might
also have a network chat were users can see what others type, much like
a CB radio.  The best systems allow conferences to be set where each
member can see what the others are typing.

Electronic mail can be a major plus in networks.  Each user should be
able to create messages, attach files, and specify by individual or groups
of individuals who is to receive the mail.  Every one should be able
to scan the mail by subject, sender, and date, and select items to look
at in detail.  They should be able to compose responses while brousing
their mail.  Senders should be able to check whether the mail they sent
has been read or has a response.

o    Calendaring and resource scheduling.

Keeping a common calendar for all persons on the network can improve
upon having to call up each person individually to schedule meeting or
to book resources like meeting rooms.  A good calendaring system has
a security system so that some entries are private, will issue reminders,
allows schedules of different persons to be consolidated when looking
for common free blocks, and allows authorized secretaries to schedule
time.  A compact print out of daily, weekly, or monthly schedules should
be available so that people do not have to be at a computer just to check
their schedules.

o    Data base.

There is one major feature of managing data that is seldom available
on a network without special programming:  record locking.  This means
what when a person is updating a part of a data base, others are prevented
from updating that part but can still be updating other parts.  That
way several persons can be doing data entry into the same file at the
same time.  Always ask of a network what multi-user data bases run on
it that support record locking.  Some data bases go even further and
in effect provide their own multi-user protection and security system
independently of the network.  The data base program will recognize different
users and constrain their access accordingly.  So even if the network
provides inadequate protection and security, the software applications
that run on the network can compensate.


Successfully Selecting and Implementing a Network

A good network needs more than a good design:  it has to have a good
implementation, which means not only that it runs fast enough, but works
reliably and can be understood by users and supported by the people imple-
menting it.  A network with implementation problems will be an unending
headache not worth the trouble.  But how do you know it runs well before
you get it?  Call up other users.  Talk especially to users who are trying
to use the network to do tasks similar to what you want.  If you can't
find users successfully doing what you want with the software you plan
to use, be worried, because you will have to check out the network yourself.
The unfortunate truth is that there is no way a network manufacturer
can check out any more than a small fraction of the possible uses of
a network, and no network runs all software than runs on a stand alone
computer.  Above all, then, don't just put a network immediately into
production.  Test every use and application out before going into production.
Build up local expertise in your network and a support staff.  Plan for
frustrations and delays.  Carefully train users on the network.  And
have people ready to troubleshoot problems as soon as they arise.  The
time, people, and resources it will take to get a network running will
be far more expensive that the cost of the hardware and software for
the network itself.

Every capability mentioned in this article is a real and natural requirement
needed by users, and yet there is no network that has even 80% of them.
Every network will have limitations which surprise and anger you, as
well as problems in its implementation which will cause the network to
run unreliably or make software not run properly.

If all this makes selecting and implementing a network seem more complicated,
discouraging, time consuming, and expensive than you ever thought it
would be, you have learned your first lesson, because that's the way
it is.  You had better know what you are getting before you marry a network
into your work or business.  Temper your expectations, because no network
has succeeded even in simply sharing all the resources and running all
of the programs that work on stand alone computers.



A Checklist for Evaluating Networks


I    What PCs run on the network?

o    PC Jr

o    PC

o    PC XT

o    PC AT

o    PC 3270

o    PC 370

o    IBM compatibles (specify)

o    Non-IBM compatibles (specify)


II   What resources can be shared?

o    Hard disks
o    Floppy drives and electronic drives

o    Printers
          High speed impact
          Letter quality
          Laser

o    Plotters

o    Modems

o    Tape Drives
          Streamers
          9 track

o    Gateways
          To other computers
          To other networks

o    Computer (its CPU)


III  Servers

o    Specialized computer required?

o    Any work station be a server?

o    Shared resource limited to particular models?

o    Can existing resources on stand alone machines be shared?

o    Must a computer that is a server be dedicated and not used as a
     work station for running other jobs?


IV   Multi-User Protection

o    Will two users writing at the same time to the same drive ever get
     their files intertwined (cross-linked)?

o    Is passive, automatic file locking supported?

o    Can the network be set up so that several users can have write access
     to the same files, yet when one is updating it no one else can use
     it (all others are locked out)?

o    If an application writes out a work file with the same name every
     time it is used, will two users who use the same program overwrite
     each others work file?


V    Security
o    Can access to the network be restricted to an authorized list of
     users?

o    Can each user be restricted to using specific public resources on
     the network?  Can the list differ for each user?

o    Can a shared hard disk drive be made read only for network users,
     or read/write?  Can it be read only for some users and read/write
     for others?

o    Can access to public resources be limited by security level rather
     than user name?

o    Can groups of files have separate security protection based on their
     names and/or location on a drive?

o    Can user stations be automatically logged off the network after
     a specified time of inactivity?


VI   Software Applications

Is there special software available on the network which supports

o    data base management with record level locking?

o    terminal to terminal communications?

o    electronic mail?

o    calendaring and time and resource management?

What software will either not run at all on the network or not work properly?