A kind of Metadata Prefetch Method for Distributed File System
Jingyi Zhang, Bo Jiang · 2021
Distributed file system has the ability to efficiently store and organize computer data, and its performance is affected by the performance of system to operate metadata greatly. This paper introduces a metadata prefetching method called “HR-Meta” which based on implicit data relevant features. This method first stitches multiple implicit features that extracted from files into feature vectors, and then makes a comparative analysis on them. Finally, the system prefetches the metadata of the file according to the model results, which improves the prefetching accuracy and cache hit ratio of the file system, shortens the access process of its related file metadata. Therefore, the file system obtains more efficient data processing capability. The simulation results show that under the ARC based replacement strategy, when the file relevance degree increase from 0.25 to 1, the cache hit ratio of the system with HR-Meta prefetching method can be increased by 9.1% to 47.7% compared with that before.