Performance optimization of microkernel-based embedded I/O virtualization techniques
Mengdi Wu, Zhe Duan, Jian Qing Li · 2024
The introduction of the virtualization layer in multi-core embedded devices leads to a semantic gap between physical hardware and applications, which causes significant delays in the processing of I/O requests in real-time domain applications. Based on a new Type-1 hypervisor—Revmm, with the microkernel architecture, this paper uses semi-virtualization to optimize the processing of I/O requests. To reduce the performance loss caused by VM switching and virtualization processing, I/O requests are aggregated and then processed in bulk. The introduction of real-time information for real-time I/O requests enables the description of the request's anticipated processing timeframe. The front-end driver employs a priority-based sorting mechanism to categorize stored I/O requests according to the real-time information, and gives priority to the I/O requests with high priority. Experimental results demonstrate that the separated embedded I/O virtualization framework based on microkernel architecture designed in this paper can reduce the additional performance overhead that occurs during virtualized I/O processing.