Survey on Use of AI/ML Algorithms in Enhancing Filesystem Performance and Efficiency

The processing power of CPUs is increasing day by day, but systems often struggle to fully utilize this power due to the bottleneck of file access time from the disk. Over the years, various techniques have been devised to reduce I/O latency while considering various file parameters. Our survey paper aims to shed light on these techniques, efficiently comparing them and highlighting the benefits and drawbacks of each approach. For our evaluation, we have considered various file parameters, such as access time, file size, directory structure, and kernel execution flow. We have combined these parameters with a range of supervised and unsupervised machine learning algorithms, including Decision Tree, ABLE, RNN, GNN, CART, XGBoost, CNN, and more. This survey aims to highlight the use of ML/AI algorithms in the intelligent detection of file access and I/O patterns to optimize the performance of modern applications.