Software restructuring in performance-critical, distributed, real-time systems

Abstract

The availability of low-cost microprocessor components is providing an expansion of multimicroprocessor applications to resolve problems where time-critical needs exist. Typically, dedicated and specialized multiprocessor systems are developed to fit the needs of such areas as robotics, nondestructive assembly-line testing, and defense systems. A difficulty in this area, and the focus of this work, is the development of the software structure of the tasks and their distribution over the network of processors such that as designer may achieve a desirable system performance. This dissertation work develops a methodology for restructuring a set of subtasks, with precedence constraints and stochastic properties, into alternative structures that can be scheduled effectively in parallel and distributed environments. Specifically, the pipelined and limited-connectivity architectures are considered. These alternative structures and their properties may be considered by a designer to select the best alternative. In addition, this methodology includes an approach for loop structuring and a technique to control the performance parameters of bottleneck modules. A queuing model, including both communication and synchronization delays, is used for analysis. A prototype software design tool is implemented to evaluate the results of the developed methodology.