In this paper, we present model-based computing techniques for the generation and evaluation of designs. Model-based computing employs multi-use declarative and executable machine descriptions to derive information from which machine software can be constructed automatically. We extend an approach for optimal scheduling of reprographic machines [Fromherz&Carlson 94] to routine system design. Parameterized machine models, in conjunction with cost and design constraints, and a scheduling algorithm are used to determine values for machine model parameters that optimize the productivity of the machine.
In the absence of precise knowledge of the application usage (execution scenario, workload distribution, etc.), design is targeted at an estimation of the "average" user's application context. We present a method for classifying optimal sets of machine parameter values with respect to workload distribution, which allows tuning design parameters dynamically for specific user requirements.
Back to the top.