| Current file systems waste a lot of disk space and perform poorly when they are directly used for storing small files. It has been a trend to store file in a centralized and distributed way. Unfortunately, current file systems can not meet the rapidly growing demands. So, this paper presents a novel file system-MiniFS. MiniFS modifies the storing structure, introduces BTree Index and Cache. Meanwhile, MiniFS creates Incremental Journalling System based on its atomic operation property. In the implementation of MiniFS, thread pool has been largely used to meet the high concurrency access need.This paper first describes the background and the requirement of MiniFS, then analyses current file systems and points out their flaws. In the second charpter, it describes the design considerations of MiniFS from the viewpoint of exploiting what current file systems do well to avoid what they do poorly. The discussion of the implementation of MiniFS comes in the following chapter. It mainly focuses on key components of MiniFS, and illustrates the greatest difficulties in the implementation of MiniFS. This paper also provides a performance test which compares MiniFS and Ext3 according to some benchmarks. At the end of this paper, the flaws of MiniFS are mentioned, and some improvements are put forward to. Also, the development direction of MiniFS is pointed out.MiniFS is a file system which is designed and implemented for storing greate number of small files. MiniFS has fully met the design goals in the utililzation of disk space and reading&writing performance by some optimizations. |
