Methods for Scheduling of xtUML State Machines Executing in Parallel

Sammanfattning

This thesis examines ways of multithreading Finite State Machines (FSMs) derived from Executable Translatable Unified Modeling Language (xtUML) models using a model compiler. The problem has little previous research done. Delimiting the problem by requiring FSMs to be Directed Acyclic Graphs and interpreting FSMs as tasks, the problem resembles a scheduling problem. With the restrictions of FSMs and model compilation taken in consideration, the critical path of an FSM is a reasonable parameter to use when scheduling. The evaluation shows that letting critical paths be static for the entire FSM yields shorter execution time compared to calculating the critical path for each individual state. This thesis also improves on the previous multithreading solution at SAAB by letting resource access conflicts be mapped between individual states instead of entire FSMs. This increased granularity in multithreading which decreased the average execution time significantly.