Abstract:
Energy management is very important and sometimes critical for certain classes of hard real-time systems. In this paper, we present effective energy reduction techniques for hard real-time systems developed in Java, which execute on bare metal and run on a time-predictable specialized Java processor. We modified traditional clock gating and dynamic frequency scaling methods to include the hardware-based run-time slack calculation in periodic tasks, thus reducing energy consumption in hard real-time systems. Two methods for energy reduction are employed leading to Energy Aware Java Optimized Processor (EAJOP). The first method includes task execution time monitoring and comparison with the estimated worst-case execution time to calculate the slack and bringing the processor to sleep for the slack duration upon task completion. The second method introduces real-time residual slack calculation at so-called checkpoints inside the periodic task, which are then used to lower the system frequency of the rest of the task dynamically, resulting in lower energy consumption. We compare EAJOP with baseline JOP when implemented on FPGA and demonstrate gains in energy consumption.
