other systems, different from the one in Figure 6, that also did the same job. We might have provided two transducers and no memory, each number remaining available at its transducer until ready to be used by the adder. We might have provided for both numbers to have gone into the memory, insulating the adder entirely from the input 'process. With such a simple task there are only a limited number of non-foolish ways to do it. Still, there are several -- enough to show that the statement of the task to be done by the system is not to be confused with the description of the system that does the job.

The desired characteristics of a system are specified in two different ways: (1) by overall objectives on the system behavior and structure; and (2) by detailed specifications on the information processing to be accomplished. Both are important and are used jointly to state what is desired. We take up each in turn.

OBJECTIVES AND EVALUATIONS

Objectives can often be stated as maximizing or minimizing some measure on a system. A system should be as reliable as possible, as cheap as possible, as small as possible, as fast as possible, as general as possible, as simple as possible, as easy to construct and debug as possible, as easy to maintain as possible -- and so on, if there are any system virtues that we have left out.

There are two deficiencies with such an enumeration. First, one cannot, in general, maximize all these aspects at once. The fastest system is not the cheapest system. Neither is the most reliable. The most general system is not the simplest. The easiest to construct is not the smallest, and so on. Thus; the objectives for a system must be traded off against each other. More of one is less of another and one must decide which of all these desirables one wasn't most and to what degree.(11)

The second deficiency is that each of these objectives is not so objective as it rooks. Each must be measured, and for complex systems there is no single satisfactory measurement. Even for something as standardized as costs there are difficulties. Is it the cost of the materials -- the components? Do we use a listed retail cost or a negotiated cost based on volume order? What about the cost of assembly? And should this be measured for the first item to be built, or for subsequent items if there are to be several? What ab6ut the costs of design? That is particularly tricky, since the act of designing to minimize costs itself costs money. What about cost measured in the time to produce the equipment? What about the cost of revising the design if it isn't right; this is a cost that may or may not occur. How do we assign overhead or indirect costs? And so on. In a completely particular situation one can imagine an omniscient designer knowing exactly which of these costs count and being able to put dollar figures on each to reduce them all to a common denominator. In fact, none of us knows that much about the world we live in and what we care about.

The dilemma is real: there is no reducing the evaluation of performance in

-------------------------

11. Belief in the perversity of nature that forces all good things to trade off against each other should not be carried to extremes. The field of digital systems provides a counter-example. Digital systems now, compared to digital systems even a few years ago, are: faster, cheaper, smaller, more reliable, simpler, and easier to maintain. In short, they are better in every way, and nothing had to be traded off in the final performing system to. obtain them. Within this, there always exist small trade offs, e.g., between speed and cost. But this is barely significant against the major trend.

23