Linux Storage IO Important Parameters Filtering Model Based on Random Forest
Zhangpin Sun, Lijun Chen · 2021 16th International Conference on Intelligent Systems and Knowledge Engineering (ISKE) · 2021
There are many key configuration parameters in the Linux storage IO module that affect the performance of the server system. The parameters will affect each other in a complicated way, and it becomes challenging to adjust the parameters to obtain high performance manually. The emergence of an automatic tuning framework alleviates this problem, but there are many parameters in storage IO, and the automatic tuning framework is usually based on machine learning algorithms. It is not efficient to explore such a vast parameter space. It is necessary to repeat the work after replacing the system. We believe that evaluating the utility of the parameters in the storage IO and pre-filter important parameters in advance for automatic tuning is necessary. The important parameters filtering method based on the random forest algorithm to filter the storage IO parameters shows that only the important parameters tuning can almost achieve the high performance obtained by all parameter tuning. In addition, the operating system is close to the hardware. Compared with the parameter optimization of the application layer program, the optimization of the operating system storage IO parameters can fundamentally maximize the IO performance of the server.