Computer Design
A computer aided design and VLSI approach

Paul J. Drongowski


Chapter 1 - The marketplace.

A computer system is a commodity to be produced and sold
to customers with specific computational needs. The
manufacturer expects to make a profit from the sale and
to support that product over its installed lifetime.
This "aftermarket" support includes hardware and
software maintenance, training and sometime operational
responsibilities. The customer expects the system to
satisfy a set of requirements within the budget allocated
for computing resources including its lifetime operational
costs.

This chapter examines the nature of the transaction between
the manufacturer and customer. It begins by describing the
needs of the customer and the demands of the marketplace.
Next, computer system requirements are presented from the
perspective of the manufacturer. How these demands and
requirements affect the technical aspects of computer
system design, test, production and maintenance are discussed
in the the final section.

Section 1 - Customer requirements.

In order for a product to be profitable, it must satisfy a
need. A customer will purchase a computer system if the
perceived need or cost of not purchasing the system become
great enough. Four different kinds of resources are typically
required to fulfill customer expectations: computation,
information storage and retrieval, communication, and
interaction with the real world.

Above all, we think of computer systems as "data manipulation
engines." Given some information which is represented electronically,
we want to compute or derive new results from existing information.
The results must be produced in a timely manner as most information
has time value -- old information is usually worthless. Thus,
the customer requires one or more computational resources that will
manipulate the information of interest to that customer within
the alloted or specified benchmark time. This need may be satisfied
by one or more processing units (processors.)

Information has some permanence and must be stored until the
value of the information is lost. In computer system design,
at least three different informational lifetimes may be noted
with different store sizes and retrieval (access) times.
While data is being manipulated, many intermediate results may
be produced and discarded. The life of this kind of information
is short, but typically requires high speed access for fast
computation. Short duration, fast access information is usually
maintained in the general registers, processor cache or primary
memory of the computer system. Computation produces information
with an intermediate lifetime. This information is stored between
periods of computation and requires moderate to very large
storage elements such as magnetic disks. Finally, long term
information must often be maintained for historical or archival
purposes. Access to the information is infrequent, but requires
the least expensive storage mechanism possible due to very
large volume of data that must be retained. Magnetic tape and
digital compact disks (CD-ROM) are useful here.

Communication is an essential element in any computing system.
At the very least, communication channels are needed to move
information from storage to computational elements and back again.
Many applications such as banking, military exercises and
government depend upon high speed, reliable computer to computer
telecommunications.

A specialized form of communication exists between the computing
system and "the real world." Human to computer communication
belongs in this category. Humans are used to the high bandwidth
physiologically adapted forms of communication such as speech
or handwriting and often become frustrated with the "typewriter"
style of most human and computer interactions. Thus, graphics,
speech and vision processing and other forms of analog and digital
signals will become increasingly important to the usability of
computing systems. Since real world phenomena are continuous
or analog in nature, conversion to and from the discrete domain
of digital computation is required.

Given the availability of these resources, the customer also
expects the tools to control and manage their use. The electronic
devices that compute, store, retrieve and communicate information
must be made into readily usable resources such as software
processes (programs), files, databases, graphic renderings and
intercomputer messages. This is usually the responsibility of the
operating system, programming environment, compilers and packaged
(turnkey) application software.

Section 2 - Market considerations.

The sophistication of modern operating systems and compilers has
radically changed the customers view of computing systems in
general. Before the popularity of higher level programming
languages, the features the machine instruction set were critical
to the economical production, testing and use of programs.
Today, however, the typical user is far removed from the "bare
machine." A spreadsheet user is not concerned about the addressing
modes or general register set. However, the user has definite ideas
about convenience, speed of computation, size and presentation
of the spreadsheet and the availability (reliability) of the
tool. Software developers and system integration shops are interested
in aggregate performance, too, not just the end users. Questions
here are "Is it a good C (Pascal, Ada, Fortran, LISP, Prolog)
machine?" or "Can a process respond to an external event within
30 milliseconds?" The answers depend upon the efficiency of the
operating system or the ability of the compiler to produce tight
machine code as well as the base speed of the machine hardware.

The demands of the user go beyond the technological achievement
of the engineering team in the backroom. The user knows her
problem and wants to buy a set of information processing resources
to solve that problem. Customer requirements may be stated in
terms of a benchmark program or problem which the computing system
must solve. As most customers are price sensitive, they want to
buy the lowest cost product that satisfies the need. Finally,
a product solution that is available today is worth far more
in sales than a product which is unavailable or a late entry.

With a sizable investment in application software, a customer
may desire compatibility with a particular instruction set and
operating system such that the machine code may execute on the
"compatible." The standardization of programming languages,
operating systems, graphics packages and communication protocols
helps to alleviate this problem provided that the source code
for the application is available. The user (or developer) may
then "port" the package to a new host machine without compatibility
concerns.

One further complication to the marketplace is the variety and
range of computing applications. It is difficult and economically
inappropriate for one machine to satisfy all users. Applications
such as text processing and spreadsheets do not require machine
resources beyond the ordinary general purpose computer. Military
applications, however, often have very specialized requirements --
features that the business user does not want and will not purchase.
Applications that can be supported by general purpose machines
are "high volume" since a single machine can be produced in large
numbers and employed by many users for perhaps dissimilar uses.
Specialized machines are usually "low volume" with only one or a
few applications and users.

Section 3 - The manufacturer.

The manufacturer must translate broad problem-oriented user
requirements into technologically objective criteria for the
design of the computing product. The market is usual broken into
"market segments" which are customers and problems with similar
requirements. As a corporate strategy, a manufacturer may target
its system toward a particular segment or "niche." As we will
see, the size and needs of the segment will directly affect the
price at which the product may be offered for sale.

Computational requirements may be stated in terms of several
benchmark problems or throughput requirements. The problems are
solicited from customers in the target segment so that the
benchmarks will reflect the needs of real users. A benchmark
specification may state that "Machine M must execute program P
in T clock time units, use B bytes of primary memory and S bytes
of secondary storage." A less useful (but often quoted) measure 
of performance is the number of operations performed per second.
Computer designers will often speak of thousands of
instructions per second (KIPS), millions of instructions per
second (MIPS), floating operations per second (FLOPS) or even
logical inferences per second (LIPS.) This measure does not
provide any guarantee that a particular application will be
executed with sufficient speed as to be timely.

The manufacturer must determine the price that customers are
willing to pay for a particular level of performance and when
a machine of that performance level must be available for
delivery. The expected number of units to be sold (demand)
must also be estimated. These decision factors are found through
market surveys and customer contact.

Section 4 - Technical challenge.

Once the issues of performance, price, time to market and
volume have been settled, development may begin. The selling
price of the system may be separated into several components.

  * The most important component is profit. The owners
    (shareholders) demand profitability so that the equity
    value of the corporate stock will grow and that
    dividends may be maximized.
  * If software is "bundled" with the hardware, a portion of
    the development cost of the software must be included in
    the selling price.
  * Of course, the cost of producing the unit must be covered
    by the asking price. This is called the "recurring cost"
    because it will occur again and again as each unit leaves
    the production line.
  * A portion of the development cost of the hardware must also
    be added into the unit price.

The latter component is a portion of the "non-recurring cost" of
the system. The non-recurring cost includes any and all development
costs such as managerial, engineering and clerical salaries, support
services and the cost of building and testing prototypes.

The recurring and non-recurring costs must be kept low in order to
keep the selling price down or to increase the profit margin. Recurring
costs may be reduced by using fewer components or modules thereby
decreasing the physical size, power supply and cooling requirements
of the system. Simple mechanical design and attention to manufacturing
concerns can play a significant role in cost reduction. Expensive
human labor charges can be decreased by raising the level of
automation on the assembly line. The involvement of production personnel
in all stages of the design process can have a beneficial effect on
profitability and competitiveness.

Non-recurring costs can be reduced by increasing the productivity of
the engineering team through computer aided design (CAD.) Modeling,
schematic capture, simulation, place and route, design rule checking
and mechanical design aids help the engineering team to manage the
enormous complexity of system design and improve the quality of design
decisions. CAD tools may also help compress the project schedule
thereby affecting the time to market for the product.

The non-recurring cost and expected sales volume for the system have
a substantial effect on the selling price. The portion of the non-recurring
costs which is included in the price is equal to the total development
costs divided by the number of units to be sold.

                                   Total non-recurring cost
     Non-recurring cost per unit = ------------------------
                                       Estimated volume

Systems to be produced in large quantities can be sold at a lower
price since the development cost can be "amortized" across a large
number of units. Specialized, low volume products will be more
expensive (assuming equivalent development costs.) If the non-recurring
cost is underestimated or the volume is overestimated, profitability
will be in jeopardy.

To be a successful product, deliverable units must be available
when the market is ready -- the so-called "market window" for the
product. Quite often, the engineering team must take a conservative
approach to the design of the system to reduce the risk of missing
the market window. Conservative design dictates the use of proven
methodologies, tools and technology. Thus, the technology available
at a particular time has a substantial effect upon the kinds of
machines which may be successfully constructed and sold.

Computer vendors have discovered that after market maintenance
support is a lucrative source of income. The complexity and
proprietary nature of software and hardware components often
forces a customer to arrange a maintenance contract with the
vendor. Monthly charges are typically one percent of the purchase
price of the system. As this is a fixed source of income for
the vendor, the return from maintenance contracts can be improved
by increasing the reliability of the system and thus reducing
the cost of actual field service and repair operations. Increased
reliability incidentally improves customer relations and enhance
the reputation of the product line.

The challenge to computer designers, therefore, is to produce a computer
system design which satisfies the functional and budgetary constraints of
its potential customers within a market window while returning a good
profit to the manufacturer. A good design must take manufacturing and
maintenance into account to guarantee a competitive position in the
marketplace and profitability throughout the product life cycle.

Copyright (c) 1987-2013 Paul J. Drongowski