Memory-Aware Dynamic Frequency Scaling For Saving Energy In Real-Time Systems
In this study, a Memory-Aware Dynamic Frequency Scaling (MDFS) mechanism is proposed to efficiently reduce energy consumption in a real-time system. To minimize energy consumption and meet the deadline of a time-constrained application, a mathematical model is proposed to show at which frequency a processor should be run to reduce the maximum energy consumption when the cache miss rate and the relax time of the application are given. Accordingly, the proposed MDFS mechanism predicts the number of potential execution instructions of an application with the cache miss rate to evaluate the proper frequency of a processor to save energy. In addition, to guarantee the deadline requirement of an application, a relax time timer (RTT) is used in the proposed MDFS mechanism. The simulation results showed that our proposed MDFS mechanism and RTT consumed less approximately 22% of energy than the processor that used the highest frequency, and satisfied the real-time requirement for time-constrained applications.
Index Terms - Dynamic voltage and frequency scaling, unavailable time, cache miss, real-time applications.