Multi-shape tasks scheduling for online multitasking on FPGAs

Reconfigurable system on Chip (RSoC) is seen as a possible future trend in embedded electronic. This work focuses on the management of dynamically reconfigurable resources (eg. FPGA) using a real time operating system (RTOS) especially designed for such platform. Scheduler and placer are crucial OS services to allow for online management of the computing resources. In this work we introduce multi-shape task management. In multi-shape tasks management the system can choose between several versions of the same task featuring different shapes, sizes and execution times, in order to optimize resource usage. In this paper we present a comparison of several scenarios of a scheduling/placement algorithm for hardware tasks on FPGA using different number of shapes per task. Studied cases have been implemented in C++ and compared regarding runtime overhead or execution time and system efficiency. We show that, compared to a reference case as single-shape EDF (earliest deadline first) algorithm, the FPGA utilization ratio is significantly improved and the task rejection ratio is halved by using only two shapes per task.