File Type Classification for Adaptive Object File System

This paper gives an overview of a novel storage management concept, called, adaptive object file system (AOFS). The design of the file type classification module in AOFS is emphasized. The design attempts to increase the efficiency through the dynamic tuning technique, which automatically classifies files using attributes and access pattern. The file classification, thus, allows files to be stored in the most efficient way. The key idea is to utilize the file properties, such as access pattern, owner, size, and permissions to select adaptive file system policies (e.g. disk allocation, redundancy, and caching strategies). Moreover, a metadata store is maintained to provide the best possible dynamic tuning strategy for any given operating period. The static classification initial design is done based on decision tree, while the dynamic classification adapts the Hidden Markov model for prediction. The main goal of the AOFS design is to enhance system performance, storage efficiency and flexibility