摘要

Readahead is an important technology for improving IO performances Its performance rests with the prefetching policy We introduced a Markov decision model to describe low-level read processes and readahead behaviors such as those in modern operating system kernels when there is enough memory for caching By using this model, it is possible to analyze various readaheads and find better prefetching policies for specific read patterns with fewer practical tests For illustration, we presented an example about the ondemand readahead in linux kernel The experiments show the model agrees with the real readahead and the found policies significantly outperform the current in real world