Knowledge-based performance evaluation of computer architectures

摘要

Presents an application area for artificial intelligence techniques in the context of designing and evaluating computer architectures. A knowledge-based approach to performance evaluation of computer architectures is described which is suited for application during the early phases of design. Taxonomical knowledge is used to develop an architecture description facility that allows descriptions to be constructed quickly and easily. Heuristical knowledge about the influence of design characteristics on the architecture's performance helps to assess the run time for benchmark programs with respect to a specific architecture, without the need for time-consuming simulations. Using these kinds of knowledge, a system can be built which is flexible and fast enough to allow the comparison of many design alternatives during the early phases of architectural design. A prototypical system has been implemented; and it is concluded that the results obtained by this system support the feasibility of the knowledge-based approach to performance evaluation.