Linux Storage I/O Performance Optimization Based on Machine Learning

With the rapid development and use of high-concurrency applications, the I/O performance requirements of Linux storage systems are getting higher and higher. In the Linux storage system, unreasonable I/O parameter settings often affect performance, and it is extremely challenging to manually adjust parameters to improve I/O performance. This paper analyzes the Linux storage I/O protocol stack, selects the I/O scheduling strategy with the best performance in a high concurrency environment and the parameters that affect performance, and proposes a machine that combines the XGBoost prediction algorithm with Bayesian optimization Learn the model, apply it to the multi-parameter adjustment of the I/O scheduling strategy, recommend the most suitable parameter combination in the actual scenario, and exert the maximum performance of the I/O scheduling strategy. The experimental results show that the improved parameter recommendation algorithm greatly improves the Linux storage I/O performance in high concurrency scenarios.