When computers first appeared, input-output commands were minor afterthoughts to cohesive, often well crafted and occasionally pretentious programming languages. Today, these commands occupy over 70 percent of a programming system's instructions. Yet they, along with the user interface structures that they define, are far from cohesive, and, at least up until now, immune to standardization. We must therefore turn our thinking around and create a new breed of programming languages that are first and foremost input-output oriented and that integrate traditional processing commands into new user-oriented structures. And just as we know today that traditional commands fall into a handful of fixed categories - decision, repetition, naming, procedure definition and use — we need to search for and identify the corresponding natural classes of commands for user interfaces.

The problems of hard-real-time task scheduling in a multiprocessor environment are discussed in terms of a scheduling game representation of the problem. It is shown that optimal scheduling without a priori knowledge is impossible in the multiprocessor case even if there is no restriction on preemption owing to precedence or mutual exclusion constraints. Sufficient conditions that permit a set of tasks to be optimally scheduled at run time are derived.>

A physicomathematical basis is used to establish bounds TD(n) on the time needed to compute n-argument functions by spatially distributed primitive devices or composite systems D. The axioms used concern the speed, packing density, and noise threshold of the energy with which any computing device detects or alters the physical representation of information. The principal result is that TD(n) grows at least as n1/2. Composite systems consisting of spatially distributed identical components are examined in light of this bound. Inherent bounds on the computing time of n-argument functions are then combined with TD(n), resulting in a measure of computational efficiency which bounds computing time to processor size.

This paper is concerned with a general approach to the on-line simulation of block-diagram systems. These systems may be analog, digital, or hybrid; memory or memoryless; explicit or implicit (i.e., possessing loops of memoryless elements); vector or scalar; or of a more specialized nature, such as dynamic systems with integrators as memory elements; and simulation may be conducted in one or more dimensions, such as time and/or space.

A parametric display technique for on-line graphical output is presented. This technique is particularly useful for rapid display of continuous ``smooth'' data, through relatively few computer commands. Graphical data to be displayed are composed from primitive segments, each of which corresponds to one computer command that specifies starting and stopping point, final slope, and ``curvature.'' Presented topics include the basic principle of operation, a class of hardware realizations, typical software requirements, and experimental results of a display prototype.

The problem of deciding whether or not an arbitrary N-variable switching function is realizable with a single-threshold-element device is known to be convertible to the global minimization of a functional derivable from the given switching function. This functional is shown to consist of a structure of intersecting hyper-planes each of which is related to some threshold function. It is further shown that the functional is a convex function of its arguments so that test synthesis by minimization is a valid procedure. Two such computer-implemented minimization techniques are discussed. Finally, it is shown that this approach cannot be directly extended to threshold-element network synthesis, since there exists no convex functional with a global extremum at a network realization.

An approach based on correlation techniques is developed for the analysis and synthesis of threshold elements. A necessary and sufficient condition for linear separability is derived. Single-threshold-element realizability of arbitrary Boolean functions of n variables is shown to be dependent on a (n+1)-tuple of real numbers (characteristic vector), derivable solely from the given Boolean function. The task of finding appropriate weights for a given threshold function is converted to the minimization of a functional. Approximation to this functional relates the desired weights with the characteristic vector. As the degree of approximation increases, exact solution for the weights becomes possible at the expense of increased computational complexity. The absolute value of characteristic vector components is used to tabulate threshold functions compactly. A table is given for functions of up to six variables which permits one to determine directly, whenever it is possible, the weights and threshold that realize a given function without Boolean manipulation of that function into a standard form.