Energy Efficient QoS-Aware Random Network Coding on Smartphones

Abstract

Random network coding (RNC) technology is known to benefit various facets of information networks; however, there have been concerns for the computational complexity of RNC since its incipience. For instance, RNC’s high complexity can be directly translated into high energy consumption and drain fast smartphone batteries, making it unsuitable for mobile environments. In this paper, we optimize the energy consumption of RNC implementations with a given QoS requirement, especially throughput, for smartphone environments. To this end, we propose a duty cycling approach minimizing the energy consumption of RNC with a given throughput constraint. By manipulating the processor clock frequency controlling mechanism (a.k.a. governor) in Android, our approach can indirectly regulate the processor clock frequency and enhance energy efficiency. Real experiments on Android systems with smartphone application processors such as Samsung’s Exynos 5410, show that our method can reduce the energy consumption of RNC by up to 67% compared to a RNC implementation relying on ondemand governor for frequency control. Finally, we argue that our method can be applied to a wide range of applications by implementing it with a fast Fourier transform algorithm.

Appendix

Ondemand-sec governor is modified from the original Android ondemand governor to integrate the core cluster switching mechanism in the governor and to allow a finer-grain control over the operating frequency. Using the notion of virtual frequency, the governor can make the existence of heterogeneous core clusters oblivious to the other part of the system. That is, it appears to user applications that the governor controls the operating frequency of a core cluster up and down from 250 to 1600 MHz. If necessary, however, the operating core cluster type can be identified from the operating frequency, i.e., a frequency between 250 and 600 means that the Low cluster is operating and a frequency varying from 800 to 1600 means that the High cluster is in operational.

In the ondemand-sec governor shipped with ODROID-XU + E [25], to decide when the frequency should increase, the utilization is compared to 60% or 90% depending on if the frequency at the sampling point is under 600 MHz or over 500 MHz, respectively. When increasing, starting from the minimum value of 250 MHz, the frequency rises stepwise to 600 MHz, then to 1200 MHz, and finally to 1600 MHz. The core cluster switch (from Low to High) occurs when the frequency is increased from 600 MHz to 1200 MHz. When stepping down, the frequency reduction ratio is calculated similarly to the procesure in the Android’s basic ondemand governor and the utilization is compared to 40% or 87% if the operating frequency at the time of examination is below 800 MHz or above 600 MHz, respectively. The core cluster is switched from High to Low when the frequency is diminished from 800 to a frequency below 800. A summary of ondemand-sec governor is given in Algorithm 4.