Prof. Dr. Frank Bellosa

Thorsten Gröninger

Abstract
Energy efficiency has become a significant factor in the realm of computing, extending from battery life concerns in laptops and portable computers to data centers, where single watt differences can cost thousands of dollars in electricity.
A recent development has been the creation of heterogeneous processor architectures with high-performance and high energy-efficiency cores, e.g., Arm’s big.LITTLE or Intel’s P/E core design. To maximize the potential of these architectures, Intel has developed Thread Director, which assists operating system schedulers with hints on which core to place specific threads. In this work, we evaluate the hints given by Thread Director from an energy efficiency perspective by designing and implementing a userspace scheduler in the seL4 microkernel.
Microkernels, although often lacking in functionality compared to a monolithic kernel, make for a good candidate for measuring the impact of such new features, as these systems exhibit far less noise and interference during execution, allowing for more definitive conclusions to be drawn. Typically, they are easier for a developer to become familiar with and allow easy extension inside the kernel, allowing more rapid prototyping and testing of features once one is familiar with their features.
We evaluate our scheduler and Intel Thread Director on a desktop Arrow Lake CPU, finding that while we did not improve energy efficiency compared to a scheduler aware of heterogeneous processors, we found unusual behavior in the classification behavior of Thread Director, showing several orders of magnitude difference between class zero and other class types, which should be more thoroughly examined by futurework. Furthermore, we evaluated different methods for our userspace scheduler to sleep without kernel involvement, achieving up to a 50% uplift in energy efficiency compared to busy-waiting.

BibTex:

@bachelorthesis{thomas26EnergyEfficiency,
  author = {Nikita Thomas},
 title = {Improving Energy Efficiency with Hardware-assisted Scheduling on Heterogeneous CPUs},
  type = {Bachelor Thesis},
  year = 2026,
  month = mar# "27",
  school = {Operating Systems Group, Karlsruhe Institute of Technology (KIT), Germany}
  }