Towards OS kernel acceleration in heterogeneous systems

While Moore’s law continues to hold true, limits on the electrical distance between CPU core components have forced the maximum operating frequency of the CPU core to plateau. CPU vendors have introduced acceleration features to compensate for this frequency limit; however, these acceleration features are limited in their design because they cannot be tuned to target a specific high-level software system architecture. In this paper we evaluate the potential for an operating system kernel to be accelerated in the programmable logic part of a heterogeneous system. We consider accelerating the scheduler of the seL4 micro-kernel as a case study for feasibility. Lastly, we evaluate the proposed hardwaresoftware co-design on commercial off-the-shelf hardware and identify architectural limitations of the approach.