This paper proposes an initiative data prefetching scheme for distributed file systems in cloud computing, which allows storage servers to predict and prefetch data based on historical I/O access events without involving client machines significantly. Two prediction algorithms, chaotic time series and linear regression, are introduced to enhance I/O performance by pushing prefetched data to relevant clients, thus alleviating the burden on resource-limited machines. Evaluation experiments demonstrate that this technique effectively improves I/O performance in cloud environments.
Related topics: