Improving the performance of CPU architectures by reducing the Operating System overhead

Abstract

The predictable CPU architectures that run hard real-time tasks must be executed with isolation in order to provide a timing-analyzable execution for real-time systems. The present article discusses several solutions to improve the performance of CPU architectures and eventually overcome the Operating Systems overhead inconveniences. This paper focuses on the innovative CPU implementation named nMPRA-MT, designed for small real-time applications. This implementation uses replication and remapping techniques for the program counter, general purpose registers and pipeline registers, enabling multiple threads to share a single pipeline assembly line. In order to increase predictability, the proposed architecture partially removes the hazard situation at the expense of larger execution latency per one instruction.