Building the Case for Temperature Awareness in Energy Consumption Models: an Application of the Energy-Frequency Convexity Rule

Abstract

Optimizing computing and communication systems that host energy-critical applications is becoming a key issue for software developers. In previous work, we introduced and validated the Energy/Frequency Convexity Rule for CPU-bound benchmarks on recent ARM platforms. This rule states that there exists an optimal clock frequency that minimizes the CPU's energy consumption for non-performance-critical programs. We showed that the Energy/Frequency Convexity Rule is related to the non-linearity of power with respect to frequency and is not dependent on the supply voltage. Here, we discuss the application of an analytical energy consumption model proposed previously to our target board, a TI AM572x EVM. We show that this non-linear analytical model can, for our experimental settings, be approximated by a frequency-linear variant, as our voltage is maintained constant. This, however, does not fit the measurements on the board, suggesting that a parameter is currently missing in the analytical model. We conjecture that accounting for temperature in the model would yield more accurate results that are in-line with our measurements. This builds the case for the inclusion of this important parameter in our energy models.