Improving the System Performance by a Dynamic File Prediction Model

As the speed gap between CPU and I/O is getting wider and wider, I/O latency plays a more important role to the overall system performance than it used to be. Prefetching consecutive data blocks within individual files has been explored to hide, at least partially, the I/O latency. Unfortunately, this does not reduce the inter-file latency occurring when a program reads new files during its execution. We implemented a file prediction technique called Program-Based Successor (PBS), which effectively predicts and prefetches upcoming file requests for programs in execution. Inspired by the PBS model, we developed a dynamic prediction model called Dynamic Program-Based Successor (DPBS). DPBS dynamically adjusts the number of files prefetched according to the real-time system environment. Consequently, compared with PBS, it can further leverage the effectiveness of prefetching. We used multiple modified Andrew benchmarks to evaluate our DPBS system implemented in Linux kernel. The results show that the DPBS system can effectively reduce the elapsed time by up to 18%.