Use of model‐based architecture attributes to construct a component‐level trade space
David McKean, James D. Moreland, Steven C. Doskey · Systems Engineering · 2019
Abstract Component‐level system design first builds a design trade space of alternate solutions composed of one or more system function allocations (or partitions) to one or more physical hardware or software product component configurations. Alternate solutions are then analyzed to select a near‐optimal solution that satisfies all system requirements. The past two or more decades have seen much research into solution of the hardware/software partitioning problem. A traditional assumption has been that software executes on a single processing resource, whereas modern computer systems provide a heterogeneous combination of processing resources. These processing technologies enable performance (speed) optimization through algorithm parallelization at the expense of potentially conflicting power and thermal concerns. Current research focuses primarily on function partitioning with heterogeneous processors to achieve performance optimization. There still exists the problem of quantifying function partitioning for additional architecture attributes (electrical energy, thermal, reliability, etc.). This article presents an extensible, executable model‐based system engineering framework that supports these architecture attributes. A proof‐of‐concept executable performance attribute model for single‐CPU systems is presented that produces static performance estimates to support optimization analysis and dynamic estimates for simulation analysis. This model‐based method replaces current spreadsheet approaches for architecture analysis of alternatives. Improved performance estimation reduces the need for prototype development, saving time and money. Improved architecture attribute estimates foster early communication with specialty engineers, improving design through better informed trade‐offs. Design space exploration fidelity is improved through the ability to easily model computer resource alternatives.