Scheduler-Assisted Prefetching: Efficient Demand Paging for Embedded Systems

In an attempt to substitute NOR flash with NAND flash and provide more memory to applications, embedded systems have to use demand paging. However, demand paging drastically degrades performance when the page faults rate is high. Prefetching has been known as a common remedy for the page fault overhead. Although many prefetching mechanisms have been proposed, they are efficient only for specific page access patterns. Others tend to be either too naïve to provide impressive results or too difficult to implement. We propose the scheduler-assisted prefetching mechanism which does not have such fundamental defects. As a proof of concept, this mechanism was completely implemented in Linux. We also have conducted a series of experiments to show the effectiveness of our approach. The experimental results showed a significant improvement: the number of the major page faults and the scheduling latency decreased by 30 % and 51%, respectively. 1