Enabling an OS kernel for large data with a SIMD Unit

Abstract

Nowadays, embedded systems treats larger data than ever before. It can be expected that the size of data treated by embedded systems will be increased. In ordinary case, these complicated requirements are achieved with adopting OS(operating system) kernel to systems. To improve the performance of OS kernel's data processing is meaningful for many embedded solutions. To achieve this improvement, we focused on a SIMD(Single Instruction Multiple Data) unit. Utilization of SIMD unit enables systems to treats large data effectively. Many processors include a SIMD unit for power-efficient processing. Many embedded products includes microprocessors including SIMD unit. Usually the OS kernel abandons using of a SIMD unit, but there are places where a SIMD unit works effectively. There are possibilities that improve performance of data processing with a SIMD unit in the OS kernel. This paper describes experiments to explore the possibilities to utilize a SIMD unit in the kernel. We performed experiments by using UML (User Mode Linux) and increased its performance with utilizing a SIMD unit. This paper shows that a SIMD unit can improve the OS (Operating System) kernel's data copying performance.