Experimental energy profiling of energy-critical embedded applications

Abstract

Abstract-Despite recent advances that have greatly improved the performance of embedded systems, we still face many challenges with regard to energy consumption in energy-constrained embedded and communication platforms. Optimizing applications for energy consumption remains a challenge and thus is a compelling research direction, both on the practical and theoretical sides. This paper presents a new experimental bench for energy profiling of non-performance-critical embedded and mobile applications and reports preliminary results obtained on two embedded boards. The experiments are driven by an online energy monitoring mechanism using National Instruments' cDAQ and LabVIEW running on a host machine. The host monitors a target device, which runs a set of benchmarks. We describe the experience gained from using and modding two different target boards, namely an Nvidia Jetson TX1 and a TI AM572x evaluation module. In particular, we confirm, and thus further validate, the existence of the Energy/Frequency Convexity Rule for CPU-bound benchmarks. This rule states that there exists an optimal clock frequency that minimizes the CPU energy consumption for non-performance-critical applications. We also show that the gain of frequency scaling is highly dependent on workload characteristics. Any future energy-management approach should take these behavioral traits into consideration.

Experimental Workflow for Energy and Temperature Profiling on HPC Systems

“…But, mostly they can be broadly classified into two different categories [14]. They are, works that directly measure for energy consumption using external gauges [21] or on-board sensors [7], and works that model energy, based on low-level activity derived from hardware performance counters [13], [20].…”

“…External gauges from National Instruments (like NI 9227 for current measurement and NI 9215 for voltage measurement) [21] and from Monsoon power meter [3] can be used to accurately measure hardware component-level or system-wide energy consumption. On-board power sensors like Intel's Running Average Power Limit (RAPL) [18] have also been used previously to get the energy consumption of individual hardware components.…”

“…Many of the approaches were from the perspective of the CPU's dynamic voltage and frequency scaling (DVFS) without considering the impact of software applications on the overall energy consumption [12]. Others have re-validated the existence of Energy-/Frequency convexity curve in the context of embedded systems and energy-critical software applications [21]. Efraim et al [5] have also validated the existence of an optimal frequency and voltage operational point in order to achieve minimal energy consumption on Intel ® Core processors.…”

Accurate Calibration of Power Measurements from Internal Power Sensors on NVIDIA Jetson Devices

“…Authors of [5] studied and proposed a valid method to calculate the energy consumption of Graphic Processing unit (GPU) kernel activities in NVIDIA Jetson Nano and AGX Xavier by only relying on built-in sensors. Similarly, authors of [6] conducted the power analysis and proposed a benchmark to profile the energy consumption of energy-constrained embedded applications on Jetson devices, but implementing their measurement setup is not practical for all researchers. In [7], the Jetson Nano built-in input sensor for power measurement has been compared with a real-time power measurement from an external benchmark, observing a non-negligible gap between these two readings.…”

“…Unfortunately, all these valuable studies that aim to provide energy consumption analysis at the edge with the aim of minimizing the energy consumption and edge network climate impacts or network resource management, did not have the opportunity to use actual power measurements of NVIDIA Jetson boards for more accurate energy analysis. Even though authors of [6], [7] compared the built-in sensors measurement with power data measured by an external setup for specific cases, there is a lack of a general model or tool to acquire the real power data for NVIDIA Jetson devices. An accurate estimation of the power used by these devices at runtime is key to optimizing their performance and energy efficiency and ensuring their reliable operation in a variety of applications.…”