System Synthesis Methodology

Abstract

The design of concurrent real-time embedded systems, and embedded software in particular, is a difficult task, which is hard to perform manually due to the complex consumer-producer relations, the presence of various timing constraints and the tight interaction with the underlying hardware. We present here a new approach which supports the automated implementation of concurrent behavior on a single processor (multi-tasking) (and with some additional extensions also on multiple processors). More specifically, it tackles the problem of task scheduling under real-time constraints. The processor can be either an application-specific processor or an instruction-set (i.e. programmable) processor. The approach uses as much as possible pre-ordering of the concurrent behavior under real-time constraints to minimize the run-time overhead. Consequently, a minimal application-specific multi-tasking kernel is synthesized for the software portion¹ to facilitate the coordination of the partially ordered tasks, featuring selectable process scheduling models dedicated to the needs of the application. This approach has the advantage that real-time constraints can be guaranteed and the synthesized kernel (for the software portion) will likely be significantly more simple than a generic (RT)OS solution. Section 1 below gives an overview of the system synthesis methodology and introduces the concepts used in the remainder of the text.

“... model-based techniques promise to tip the balance in favor of the system implementor ...”

—D. Hard, “Biting the Silver Bullet” [Hard 92]