A Method for Processor Frequency Management Based on determining the Intensity of Memory Accesses

Abstract

Nowadays, mobile devices are required to have high performance and low power consumption. These characteristics directly depend on the processor frequency, because at high frequency the processor is able to execute more instructions per cycle, but it has to spend more energy, and vice versa.
In this paper, a study of modern approaches to processor frequency management in the Linux operating system kernel is conducted to improve processor performance and reduce power consumption of a mobile device. A new method of processor frequency controlling is developed, which is based on solving a discrete two-criteria optimization problem. Two interdependent characteristics act as optimization criteria: performance and power consumption. The proposed method is based on the idea of analyzing the instructions of the executed task to identify cases when increasing the processor frequency is inefficient due to frequent memory accesses. The problem is solved under certain restrictions: only one process running on one processor core is considered; heterogeneous processor architectures are not considered; hardware multithreading of processor cores is not taken into account, i.e. each logical core is considered as a physical core. Based on the proposed method, software was created and tested on 11 tests from Rodinia Benchmark Suite [6]. This test suite is used to evaluate the performance of computers on different types of tasks such as general purpose computing, image processing, signal processing and many others. The conducted study showed that the use of the proposed approach reduced the power consumption of the mobile device by 7-15% on average and increased the performance by 3-5% on average compared to the performance of existing utilities in the Linux kernel.