Evaluation of Power Consumption and Application Optimization for Adaptive-Ticks Feature in Linux Kernel

Scheduler timer architecture has significant impact on operating system performance and power consumption. The current generation of Linux kernel supports multiple timer implementations, including periodic ticks, Dyntick-idle and Adaptive-ticks. Adaptive-ticks kernel offers the benefits of previous generations with additional improvement in power consumption and performance. In this paper, we evaluate the impact of Adaptive-ticks on power consumption with Linux kernel version 5.4.0 on an Intel Core i9-9900K. The current generation of Adaptive-ticks feature does not support multiple tasks in a ready queue; however, with the increase in application parallelism, not having support for multiple tasks in the ready queue poses a significant disadvantage to this feature. To support multi-threaded applications, we propose an application optimization technique which splits threads into two main categories, lightweight and heavyweight, with proper affinity settings for better power consumption. In addition, this study proposes a possible implementation strategy to extend the Adaptive-ticks feature to support multiple tasks in the ready queue. Our tests use in-band “RAPL” for profiling power consumption, and synthetic benchmarks such as Livermore, RAMSpeed, and SysBench, as the workloads. For real-world application benchmarking, we use Linux kernel compilation. The study shows that Adaptive-ticks kernel can reduce power consumption by 1–2.7% and the application optimization technique provides a 2.4% enhancement in power consumption.