Schedulability-driven scratchpad memory swapping for resource-constrained real-time embedded systems

In resource-constrained real-time embedded systems, scratchpad memory (SPM) is utilized in place of cache to increase performance and enforce consistent behavior of both hard and soft real-time tasks via software-controlled SPM management techniques (SPMMTs). Real-time systems depend on time critical (hard) tasks to complete execution before their deadline times. Many real-time systems also depend on the execution of soft tasks that do not have to complete by hard deadlines. This thesis evaluates a new SPMMT that increases both worst-case task slack time (TST) and soft task processing capabilities, by combining two existing SPMMTs. The schedulability-driven ACETRB / WCETRB swapping (SDAWS) SPMMT of this thesis uses task schedulability characteristics to control the selection of either the average-case execution time reduction based (ACETRB) SPMMT or the worst-case execution time reduction based (WCETRB) SPMMT. While the literature contains examples of combined management techniques, until now there have been none that combine both WCETRB and ACETRB SPMMTs. The advantage of combining them is to achieve WCET reduction comparable to what can be achieved with the WCETRB SPMMT, while achieving significantly reduced ACET relative to the WCETRB SPMMT. Using a stripped-down RTOS and an SPMMT simulator implemented for this work, evaluated resource-constrained scenarios show a reduction in task slack time from the SDAWS SPMMT relative to the WCETRB SPMMT between 20% and 45%. However, the evaluated scenarios also conservatively show that SDAWS can reduce ACET relative to the WCETRB SPMMT by up to 60%