I/O feature-based file prefetching for multi-applications

Applications such as multimedia, date mining-and scientific computing may exhibit sophisticated data access patterns that are different from one application to another. I/O prefetching is considered to be effective in improving I/O performance. However, the efficiency depends on identification of patterns for future I/O accesses, which is difficult to achieve in multi-application environment. This paper presents the design, implementation, and evaluation of an I/O feature-based prefetching strategy (FFP) for multiple application systems. The key idea of FFP is to extract I/O features from the collected file access information to guide prefetching. We first analyzed the I/O access patterns and designed a simple and effective feature notation to represent patterns. Then we developed a toolkit to generate and update I/O features automatically. Finally, we designed and implemented an I/O feature-based file prefetching thread library for the multi-application environment. Experimental results show that the proposed prefetching mechanism (FFP) improves I/O performance significantly for applications with complex I/O data access patterns.