Can Real-Time Systems Benefit from Dynamic Partial Reconfiguration?

In real-time systems, a solution where hardware accelerators are used to implement computationally intensive tasks can be easier to analyze, in terms of worst-case execution time (WCET), than a pure software solution. However, when using FPGAs, the amount and the complexity of the hardware accelerators are limited by the resources available. Dynamic partial reconfiguration (DPR) of FPGAs can be used to overcome this limitation by replacing the accelerators that are only required for limited amounts of time with new ones. This paper investigates the potential benefits of using DPR to implement hardware accelerators in real-time systems and presents an experimental analysis of the trade-offs between hardware utilization and WCET increase due to the reconfiguration time overhead of DPR. We also investigate the trade-off between the use of multiple specialized accelerators combined with DPR instead of the use of a more general accelerator. The results show that, for computationally intensive tasks, the use of DPR can lead to a more efficient use of the FPGA, while maintaining comparable computational performance.