Can Learning-Based Hybrid DVFS Technique Adapt to Different Linux Embedded Platforms?

With the increased use of embedded platforms powered by Linux, battery life has become one of the central focuses in research and innovation. Towards this, dynamic voltage and frequency scaling (DVFS) has emerged as a powerful technique. However, reinforcement learning-based hybrid DVFS techniques for Mixed Task sets are hardly investigated, considering Linux operating systems targeted on embedded platforms. This paper illustrates an experimental analysis of an energy reduction technique using hybrid learning-based DVFS methodology for Mixed Task sets on Linux Embedded Platforms with ARM and Intel architectures. We analyze the hybrid DVFS approach using reinforcement learning to handle a Mixed Task set on Linux platforms and evaluate the algorithms on various real-time devices. Results show that we can successfully apply the extended hybrid DVFS technique on real-time Linux embedded platforms, i.e., Dell Laptop, Raspberry Pi, and Beagle Bone Platform, all focusing on energy saving. Over 3% energy savings can be achieved for a standard real-time scheduling mechanism without any penalty in application throughput. We believe that our implementation might potentially lead to a number of innovative applications that use Learning-based DVFS algorithms targeted on Linux real embedded hardware.