A plausibility-driven approach to computer architecture design

Abstract

A dilemma skirting the discipline of computer architecture concerns the plausibility of architectural designs. To counter the widely held perception that architecture is a nebulous entity is a proposal to view such designs as specifications of constraints to be satisfied. ULISES AGUERO and SUBRATA DASGUPTA THE PLAUSIBILITY PROBLEMOne of Ihe most troublesome issues surrounding the whole discipline of computer architecture concerns the plausibility of architectural design proposals. It is widely felt (if not always explicitly stated) that a description of some new architecture (or architecture component) carries little credibility until and unless the design is supported by some sort of physical implementation.One of the reasons new architectural designs are viewed with a skeptical eye is the traditionally informal nature of such proposals. The very nature of these descriptions with their resulting lack of precision makes it very difficult to assess a design objectively for its feasibility, correctness, or performance. Indeed, one can often neither confirm nor refute the design against any given set of criteria simply because of the informality, ambiguity, and omissions of relevant detail in architectural design proposals [4]. In his tract on the methodology of science, Popper [21] termed statements possessing this lack of testability or refutability as "metaphysical statements," and in this sense architectural designs may legitimately be regarded as "metaphysical proposals."Clearly, the only way by which we can assess such a proposal is by waiting until it is implemented and then transferring our critical attention to the physical system. This, in fact, underlies the empirical, exploratory view of computer design as described, for example, by Wilkes [31]. Alternatively, we can attempt to make our design proposals less metaphysical. This can be achieved in two ways. First, by u.sing a formal architecture description language (ADL), the designer is forced This work has been supporled in parl hy GranI DCR 8408750 from the Nalional Science Foundation.

Graphs and plausibility logic

“…From the seven axioms presented A1, A2, A3, A4 and A6 figure among the principles that explicitly or implicitly appear in [1]. In addition, two axioms are included A5 and A7 [6].…”

“…In this latter discipline, it is used in the implementation of computational and/or collaborative tools and in the definition of computational designs (see [1], [2], [3], [4] and [8]). In [6,7] the Theory of Plausibility is formalized at a wider level so as to explain decision-making processes in mathematical terms.…”

Experiences with goal-oriented modeling of organizational change

“…Formulating the problem amounts, to a great deal, to solving it. To this end, one needs to develop hypotheses as to the nature of the solution and subject these hypotheses to evaluation in order to gain confidence as to their validity [25], [26].…”

Understanding design: Artificial intelligence as an explanatory paradigm

“…The fact that the engineer and the researcher in the artificial sciences are concerned with the effecting of artifacts intended to serve some purpose and that purposiveness is totally at odds with the natural sciences hardly seemed relevant (according to conventional wisdom) as far as the intellectual foundations of the artificial sciences were concerned. Since the 1960s, several works have appeared which, in one way or another, have all been dedicated to the proposition that the world of the artificial contains its own logic which is related to but is quite distinct from the logic of the natural world (Jones & Thornley 1963;Pye 1964;Jones 1980, Cross 1984Aguero & Dasgupta .1987, Brown & Chandrasekaran 1989, Coyne et al 1990, Dasgupta 1991. It has also come to be explicitly recognised that while there are many distinct artificial sciencescivil, mechanical, chemical and electrical engineering, metallurgy, aerospace technology, agriculture, computer science, organisation theory, economic and social planning, architecture etc.…”

“…may propose (whether to be performed by 'cognitive' or 'computational' agents) must take into account the fact that the state of belief about the overall design must be constantly revised as and when new evidence is invoked or the design itself evolves or changes over time. This fact -that the designer needs to constantly revise his or her claim about the design and maintain consistency amongst the belief states pertaining to different components of the design -was recognised explicitly and formed a central element of one practical design approach called the theory of plausible designs (TPD) developed by Dasgupta and his collaborators (Aguero & Dasgupta 1987;Hooton et al 1988;Dasgupta 1989bDasgupta , 1991. In fact, this work demonstrated quite dearly that the need for belief state revision during design is so immediate that automating this aspect of the design process is virtually imperative.…”