Input/output processing using on-chip memory for Cyber Physical System
Hoang Nguyen, Hakduran Koc · 2017
In recent years, most embedded systems have been extended to be able to communicate with other systems and the real-world. This new trend caused traditional embedded systems to evolve into a new generation called Cyber Physical Systems (CPS) which monitor and control complex real-world environment in which communication, computation cost and real-time constraints could be major challenges. In addition, the improvements in multi-core processors enhanced the computational power and leveraged the data parallelism. However, overall performance of the whole system is still limited by the overhead of the data transfers between sensors/actuators and computational unit. The computational unit can execute the control algorithm in a few microseconds while it may take tens of microseconds to copy dataset between sensor/actuators and computation cores. This paper presents an I/O processing scheme and DMA (direct memory access) data prefetching optimization technique to accelerate the I/O access time efficiently through on-chip memory. The proposed I/O processing scheme allows sensors/actuators to transfer data directly to and from on-chip memory. DMA prefetching optimization technique helps to prefetch I/O data before tasks executions. The experiment results show that our proposal approach reduces the execution time of the system around 26% for automatically-generated task graphs and around 21% for the task graphs extracted from benchmarks compared to traditional I/O processing schemes.