Modeling the Power Consumption of Function-Level Code Relocation for Low-Power Embedded Systems

Abstract

The problems associated with the battery life of embedded systems were addressed by focusing on memory components that are heterogeneous and are known to meaningfully affect the power consumption and have not been fully exploited thus far. Our study establishes a model that predicts and orders the efficiency of function-level code relocation. This is based on extensive code profiling that was performed on an actual system to discover the impact and was achieved by using function-level code relocation between the different types of memory, i.e., flash memory and static RAM, to reduce the power consumption. This was accomplished by grouping the assembly instructions to evaluate the distinctive power reduction efficiency depending on function code placement. As a result of the profiling, the efficiency of the function-level code relocation was the lowest at 11.517% for the branch and control groups and the highest at 12.623% for the data processing group. Further, we propose a prior relocation-scoring model to estimate the effective relocation order among functions in a program. To demonstrate the effectiveness of the proposed model, benchmarks in the MiBench benchmark suite were selected as case studies. The experimental results are consistent in terms of the scored outputs produced by the proposed model and measured power reduction efficiencies. the energy consumption of specific components, i.e., peripherals other than the microprocessor, such as the main, volatile, and non-volatile memory of the embedded system.The first category, DVFS, is a technology that varies the voltage or frequency of a computing system according to its performance and power requirements. The technique reduces energy consumption by regulating the voltage supplied to the processors of most computer systems comprising CMOS circuits. Because the power consumed is proportional to the square of the supply voltage, reducing the voltage is a highly effective way to reduce energy consumption. Some commercial microprocessors support DVFS technology to reduce power consumption. Hua et al. studied methods that optimized the voltage level to implement DVFS on a multi-voltage system to save energy in embedded systems. However, the support of multiple voltage levels results in multiple types of overhead, such as power and transition overhead. Solving these problems associated with a multi-voltage system in which such constraints exist would require determining the number of levels and values that would need to be implemented in a multi-voltage system to maximize energy saving. The ideal number of voltage levels that would allow the system to be operated in an energy-efficient manner when compared to an ideal system, that is, a system of which the voltage could be varied randomly, was experimentally proven to be either three voltage levels (2.4 V, 3.0 V, and 5.0 V) or four voltage levels (1.5 V, 2.4 V, 3.3 V and 5.0 V) [1]. Quan et al. proposed two DVFS algorithms for energy saving in a real-time embedded system. The first algorithm determines the minim...

Energy Harvesting Techniques for Internet of Things (IoT)

“…Several strategies have been proposed for prolonging the lifetime of wireless IoT devices, including duty-cycling [104], sleep scheduling [105], the reduction of the required transmission distance for IoT devices through efficient clustering [106], optimized strategies for adaptively setting the rates of sensor reading and data transmission depending on available energy [107], or the development of scheduling schemes that take into account power consumption when waking up the wireless sensing systems [108]. Some practical techniques for reducing the energy required by an IoT device, also used in the development of low-power embedded systems, include Dynamic Voltage Scaling (DVS) [109], the reduction of the frequency of the processing unit [110], or the appropriate selection of peripherals in the device [111] or of the type of memory involved in data processing and storage (i.e., Flash or RAM -Random access memory) [112], the adaptation of transmission power depending on required communication range and environment [113,114], or logic for deciding the moment and format for sending slow varying data [115 -117]. All these approaches for assuring low-power operation must be scheduled and further modified depending also on the amount of energy that is generated or stored by the IoT device at each moment in time [50].…”