Utilizing concurrency to gain performance in an industrial automation system

Sammanfattning: This work presents and discusses the results from a study, focused on achieving more performance for an industrial real-time control system. The real-time control system is used to protect electrical power stations from being destroyed by strokes of lightning. Sensors in the system continuously collect information on currents and voltages from the electrical power station which the control system protects. The sensors deliver the collected data to a computer system that bases its decisions on the arriving data. When a dangerous situation is detected circuit breakers decouple the hazardous power line.Today, the computer system is based on a single processor architecture. The problem is that this architecture does not provide enough performance to support demanding system configurations such as more advanced application algorithms and increased amount of data collected from the sensors. In order to obtain correct, timely execution of the protection applications, designers may need to optimize application code aggressively. Unwanted simplifications of algorithms or low sampling frequencies of sensor data may be the result.The motivation of this work is to study how the real-time control system is affected by being adapted to a multiprocessor or distributed architecture in order to increase the available computing resources. The objective is to improve the performance of system components in general and application components in particular. By identifying components in the existing control system that exhibit a large amount of concurrency and a relatively small amount of data exchange the study found a performance improving solution. The I/O system that is responsible for collecting sensor data and the application functionality both exhibit a large amount of mutual concurrency and may therefore scale on a system with multiple processors. In experimental configurations the I/O system components and an application model were arranged to execute in parallel on two processors. This approach exploits the concurrency available at the interface between the I/O system and application components. Results from measurements show that processing resources (up to 66% when compared with a single processor system configuration) can be freed for application components by utilizing this concurrency in a two processor configuration. The advantage gained is an increase in flexibility for application designers to select a multiprocessor system configuration for demanding applications.While parallel architectures are used in some industrial systems, not much has been written about the possibilities and threats when legacy systems are adapted to such architectures. By describing a model of an industrial real-time control system and extending that model with a mechanism that enables multiprocessor execution, we contribute to the understanding of both the functional composition and performance issues concerning parallel execution in such industrial systems.